1484 lines
44 KiB
Go
1484 lines
44 KiB
Go
package runner
|
||
|
||
import (
|
||
"fmt"
|
||
"math"
|
||
"strconv"
|
||
"strings"
|
||
|
||
"github.com/easyai/easyai-ai-gateway/apps/api/internal/store"
|
||
)
|
||
|
||
type paramProcessContext struct {
|
||
modelCapability map[string]any
|
||
candidate store.RuntimeModelCandidate
|
||
log *parameterPreprocessingLog
|
||
aspectRatio string
|
||
resolution string
|
||
}
|
||
|
||
type paramProcessor interface {
|
||
Name() string
|
||
ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool
|
||
Process(params map[string]any, modelType string, context *paramProcessContext) bool
|
||
}
|
||
|
||
type ParamProcessorChain struct {
|
||
processors []paramProcessor
|
||
}
|
||
|
||
type parameterPreprocessResult struct {
|
||
Body map[string]any
|
||
Log parameterPreprocessingLog
|
||
}
|
||
|
||
type parameterPreprocessingLog struct {
|
||
ModelType string `json:"modelType"`
|
||
Input map[string]any `json:"actualInput"`
|
||
Output map[string]any `json:"convertedOutput"`
|
||
Changed bool `json:"changed"`
|
||
Changes []parameterPreprocessChange `json:"changes"`
|
||
Model map[string]any `json:"model,omitempty"`
|
||
}
|
||
|
||
type parameterPreprocessChange struct {
|
||
Processor string `json:"processor"`
|
||
Action string `json:"action"`
|
||
Path string `json:"path"`
|
||
Before any `json:"before"`
|
||
After any `json:"after"`
|
||
Reason string `json:"reason"`
|
||
CapabilityPath string `json:"capabilityPath,omitempty"`
|
||
CapabilityValue any `json:"capabilityValue,omitempty"`
|
||
}
|
||
|
||
func NewParamProcessorChain() ParamProcessorChain {
|
||
return ParamProcessorChain{
|
||
processors: []paramProcessor{
|
||
resolutionNormalizeProcessor{},
|
||
aspectRatioProcessor{},
|
||
contentFilterProcessor{},
|
||
inputAudioProcessor{},
|
||
durationProcessor{},
|
||
audioProcessor{},
|
||
imageCountProcessor{},
|
||
},
|
||
}
|
||
}
|
||
|
||
func preprocessRequest(kind string, body map[string]any, candidate store.RuntimeModelCandidate) map[string]any {
|
||
return preprocessRequestWithLog(kind, body, candidate).Body
|
||
}
|
||
|
||
func preprocessRequestWithLog(kind string, body map[string]any, candidate store.RuntimeModelCandidate) parameterPreprocessResult {
|
||
params := cloneMap(body)
|
||
modelType := strings.TrimSpace(candidate.ModelType)
|
||
if modelType == "" {
|
||
modelType = modelTypeFromKind(kind, params)
|
||
}
|
||
log := parameterPreprocessingLog{
|
||
ModelType: modelType,
|
||
Input: cloneMap(params),
|
||
Changes: []parameterPreprocessChange{},
|
||
Model: map[string]any{
|
||
"modelName": candidate.ModelName,
|
||
"modelAlias": candidate.ModelAlias,
|
||
"providerModelName": candidate.ProviderModelName,
|
||
"provider": candidate.Provider,
|
||
"platformId": candidate.PlatformID,
|
||
"platformModelId": candidate.PlatformModelID,
|
||
},
|
||
}
|
||
context := ¶mProcessContext{
|
||
modelCapability: effectiveModelCapability(candidate),
|
||
candidate: candidate,
|
||
log: &log,
|
||
}
|
||
if kind == "videos.generations" {
|
||
ensureVideoContent(params, context)
|
||
}
|
||
chain := NewParamProcessorChain()
|
||
processed := chain.Process(params, modelType, context)
|
||
log.Output = cloneMap(processed)
|
||
log.Changed = len(log.Changes) > 0
|
||
return parameterPreprocessResult{Body: processed, Log: log}
|
||
}
|
||
|
||
func (chain ParamProcessorChain) Process(params map[string]any, modelType string, context *paramProcessContext) map[string]any {
|
||
if params == nil {
|
||
return map[string]any{}
|
||
}
|
||
for _, processor := range chain.processors {
|
||
if !processor.ShouldProcess(params, modelType, context) {
|
||
continue
|
||
}
|
||
if !processor.Process(params, modelType, context) {
|
||
break
|
||
}
|
||
}
|
||
return params
|
||
}
|
||
|
||
func (context *paramProcessContext) recordChange(processor string, action string, path string, before any, after any, reason string, capabilityPath string, capabilityValue any) {
|
||
if context == nil || context.log == nil {
|
||
return
|
||
}
|
||
context.log.Changes = append(context.log.Changes, parameterPreprocessChange{
|
||
Processor: processor,
|
||
Action: action,
|
||
Path: path,
|
||
Before: cloneAny(before),
|
||
After: cloneAny(after),
|
||
Reason: reason,
|
||
CapabilityPath: capabilityPath,
|
||
CapabilityValue: cloneAny(capabilityValue),
|
||
})
|
||
}
|
||
|
||
func parameterPreprocessingMetrics(log parameterPreprocessingLog) map[string]any {
|
||
return map[string]any{
|
||
"parameterPreprocessingSummary": parameterPreprocessingSummary(log),
|
||
}
|
||
}
|
||
|
||
func parameterPreprocessingSummary(log parameterPreprocessingLog) map[string]any {
|
||
summary := map[string]any{
|
||
"modelType": log.ModelType,
|
||
"changed": log.Changed,
|
||
"changeCount": len(log.Changes),
|
||
}
|
||
if len(log.Changes) == 0 {
|
||
return summary
|
||
}
|
||
actions := make([]string, 0)
|
||
paths := make([]string, 0)
|
||
capabilityPaths := make([]string, 0)
|
||
for _, change := range log.Changes {
|
||
appendUniqueString(&actions, change.Action)
|
||
appendUniqueString(&paths, change.Path)
|
||
appendUniqueString(&capabilityPaths, change.CapabilityPath)
|
||
}
|
||
summary["actions"] = actions
|
||
summary["paths"] = paths
|
||
if len(capabilityPaths) > 0 {
|
||
summary["capabilityPaths"] = capabilityPaths
|
||
}
|
||
return summary
|
||
}
|
||
|
||
type resolutionNormalizeProcessor struct{}
|
||
|
||
func (resolutionNormalizeProcessor) Name() string { return "ResolutionNormalizeProcessor" }
|
||
|
||
func (resolutionNormalizeProcessor) ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
if stringFromAny(params["resolution"]) != "" {
|
||
return false
|
||
}
|
||
size := stringFromAny(params["size"])
|
||
if size == "" {
|
||
return false
|
||
}
|
||
return isImageResolution(modelType, size) || isVideoResolution(modelType, size)
|
||
}
|
||
|
||
func (resolutionNormalizeProcessor) Process(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
size := stringFromAny(params["size"])
|
||
if stringFromAny(params["resolution"]) == "" && (isImageResolution(modelType, size) || isVideoResolution(modelType, size)) {
|
||
_, capabilityValue := capabilityEvidence(context.modelCapability, modelType, "output_resolutions")
|
||
params["resolution"] = size
|
||
context.resolution = size
|
||
context.recordChange(
|
||
"ResolutionNormalizeProcessor",
|
||
"set",
|
||
"resolution",
|
||
nil,
|
||
size,
|
||
"size 使用分辨率格式，归一到 resolution 供后续能力校验和计费使用。",
|
||
capabilityPath(modelType, "output_resolutions"),
|
||
capabilityValue,
|
||
)
|
||
}
|
||
return true
|
||
}
|
||
|
||
type aspectRatioProcessor struct{}
|
||
|
||
func (aspectRatioProcessor) Name() string { return "AspectRatioProcessor" }
|
||
|
||
func (aspectRatioProcessor) ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
return modelType != "text_generate" && (stringFromAny(params["aspect_ratio"]) != "" || stringFromAny(params["size"]) != "")
|
||
}
|
||
|
||
func (aspectRatioProcessor) Process(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
capability := capabilityForType(context.modelCapability, modelType)
|
||
if capability == nil {
|
||
return true
|
||
}
|
||
|
||
aspectRatio := stringFromAny(params["aspect_ratio"])
|
||
if isEmptyParamString(aspectRatio) {
|
||
before := params["aspect_ratio"]
|
||
delete(params, "aspect_ratio")
|
||
context.aspectRatio = ""
|
||
context.recordChange(
|
||
"AspectRatioProcessor",
|
||
"remove",
|
||
"aspect_ratio",
|
||
before,
|
||
nil,
|
||
"aspect_ratio 是空值字符串，不能作为有效比例传给上游。",
|
||
"",
|
||
nil,
|
||
)
|
||
return true
|
||
}
|
||
|
||
resolution := firstNonEmptyString(stringFromAny(params["resolution"]), context.resolution)
|
||
if resolution == "" {
|
||
if values := stringListFromAny(capability["output_resolutions"]); len(values) > 0 {
|
||
resolution = values[0]
|
||
} else if size := stringFromAny(params["size"]); strings.HasSuffix(size, "K") || strings.HasSuffix(size, "p") {
|
||
resolution = size
|
||
}
|
||
}
|
||
|
||
allowed := aspectRatioAllowed(capability["aspect_ratio_allowed"], resolution)
|
||
if allowed != nil && len(allowed) == 1 && allowed[0] == "adaptive" {
|
||
before := params["aspect_ratio"]
|
||
params["aspect_ratio"] = "adaptive"
|
||
context.aspectRatio = "adaptive"
|
||
if before != "adaptive" {
|
||
context.recordChange(
|
||
"AspectRatioProcessor",
|
||
"adjust",
|
||
"aspect_ratio",
|
||
before,
|
||
"adaptive",
|
||
"模型当前分辨率只允许 adaptive 宽高比。",
|
||
capabilityPath(modelType, "aspect_ratio_allowed"),
|
||
capability["aspect_ratio_allowed"],
|
||
)
|
||
}
|
||
return true
|
||
}
|
||
if allowed != nil && len(allowed) == 0 {
|
||
before := params["aspect_ratio"]
|
||
delete(params, "aspect_ratio")
|
||
context.aspectRatio = ""
|
||
context.recordChange(
|
||
"AspectRatioProcessor",
|
||
"remove",
|
||
"aspect_ratio",
|
||
before,
|
||
nil,
|
||
"模型能力配置不允许传入任何 aspect_ratio。",
|
||
capabilityPath(modelType, "aspect_ratio_allowed"),
|
||
capability["aspect_ratio_allowed"],
|
||
)
|
||
return true
|
||
}
|
||
if aspectRatio == "" {
|
||
return true
|
||
}
|
||
if allowed == nil && validAspectRatio(aspectRatio) {
|
||
params["aspect_ratio"] = aspectRatio
|
||
context.aspectRatio = aspectRatio
|
||
return true
|
||
}
|
||
|
||
processed, ok := validateAndAdjustAspectRatio(aspectRatio, capability, allowed)
|
||
if !ok {
|
||
before := params["aspect_ratio"]
|
||
delete(params, "aspect_ratio")
|
||
context.aspectRatio = ""
|
||
context.recordChange(
|
||
"AspectRatioProcessor",
|
||
"remove",
|
||
"aspect_ratio",
|
||
before,
|
||
nil,
|
||
"传入的 aspect_ratio 不在模型允许范围内，且没有可用替代值。",
|
||
capabilityPath(modelType, "aspect_ratio_allowed"),
|
||
capability["aspect_ratio_allowed"],
|
||
)
|
||
return true
|
||
}
|
||
if processed != "" {
|
||
before := params["aspect_ratio"]
|
||
params["aspect_ratio"] = processed
|
||
context.aspectRatio = processed
|
||
if before != processed {
|
||
path := capabilityPath(modelType, "aspect_ratio_allowed")
|
||
value := capability["aspect_ratio_allowed"]
|
||
if ratioRange, ok := numberPair(capability["aspect_ratio_range"]); ok {
|
||
ratio, valid := aspectRatioNumber(aspectRatio)
|
||
if !valid || ratio < ratioRange[0] || ratio > ratioRange[1] {
|
||
path = capabilityPath(modelType, "aspect_ratio_range")
|
||
value = capability["aspect_ratio_range"]
|
||
}
|
||
}
|
||
context.recordChange(
|
||
"AspectRatioProcessor",
|
||
"adjust",
|
||
"aspect_ratio",
|
||
before,
|
||
processed,
|
||
"传入的 aspect_ratio 不符合模型能力配置，已调整为允许值。",
|
||
path,
|
||
value,
|
||
)
|
||
}
|
||
}
|
||
return true
|
||
}
|
||
|
||
type contentFilterProcessor struct{}
|
||
|
||
func (contentFilterProcessor) Name() string { return "ContentFilterProcessor" }
|
||
|
||
func (contentFilterProcessor) ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
_, ok := params["content"]
|
||
return ok
|
||
}
|
||
|
||
func (contentFilterProcessor) Process(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
content := contentItems(params["content"])
|
||
if len(content) == 0 {
|
||
return true
|
||
}
|
||
|
||
if isOmniVideoLike(context) {
|
||
filtered := filterUnsupportedOmniVideoContent(content, context)
|
||
params["content"] = mapsToAnySlice(filtered)
|
||
syncVideoConvenienceFields(params, filtered, context)
|
||
return true
|
||
}
|
||
|
||
downgradeReferenceImageIfNeeded(params, content, modelType, context)
|
||
if modelType == "video_generate" || modelType == "text_to_video" {
|
||
next := make([]map[string]any, 0, len(content))
|
||
for index, item := range content {
|
||
if isImageContent(item) {
|
||
reason, path, value := imageContentRemovalEvidence(item, modelType, context)
|
||
context.recordChange(
|
||
"ContentFilterProcessor",
|
||
"remove",
|
||
fmt.Sprintf("content[%d]", index),
|
||
item,
|
||
nil,
|
||
reason,
|
||
path,
|
||
value,
|
||
)
|
||
continue
|
||
}
|
||
next = append(next, item)
|
||
}
|
||
content = next
|
||
}
|
||
if modelType == "image_to_video" || modelType == "omni_video" || modelType == "omni" {
|
||
if !supportsFirstAndLastFrame(context.modelCapability, modelType) {
|
||
next := make([]map[string]any, 0, len(content))
|
||
for index, item := range content {
|
||
if stringFromAny(item["role"]) == "last_frame" {
|
||
context.recordChange(
|
||
"ContentFilterProcessor",
|
||
"remove",
|
||
fmt.Sprintf("content[%d]", index),
|
||
item,
|
||
nil,
|
||
"模型不支持首尾帧输入，已移除 last_frame。",
|
||
capabilityPath(modelType, "input_first_last_frame"),
|
||
map[string]any{
|
||
"input_first_last_frame": capabilityValue(context.modelCapability, modelType, "input_first_last_frame"),
|
||
"max_images_for_last_frame": capabilityValue(context.modelCapability, modelType, "max_images_for_last_frame"),
|
||
},
|
||
)
|
||
continue
|
||
}
|
||
next = append(next, item)
|
||
}
|
||
content = next
|
||
deleteFieldsWithLog(params, context, "ContentFilterProcessor", []string{"last_frame", "lastFrame"}, "模型不支持首尾帧输入，已移除快捷字段。", capabilityPath(modelType, "input_first_last_frame"), map[string]any{
|
||
"input_first_last_frame": capabilityValue(context.modelCapability, modelType, "input_first_last_frame"),
|
||
"max_images_for_last_frame": capabilityValue(context.modelCapability, modelType, "max_images_for_last_frame"),
|
||
})
|
||
}
|
||
}
|
||
params["content"] = mapsToAnySlice(content)
|
||
return true
|
||
}
|
||
|
||
func imageContentRemovalEvidence(item map[string]any, modelType string, context *paramProcessContext) (string, string, any) {
|
||
role := stringFromAny(item["role"])
|
||
switch role {
|
||
case "first_frame":
|
||
return "模型能力未开启首帧输入，已移除 first_frame。", capabilityPath(modelType, "input_first_frame"), map[string]any{
|
||
"input_first_frame": capabilityValue(context.modelCapability, modelType, "input_first_frame"),
|
||
"input_first_last_frame": capabilityValue(context.modelCapability, modelType, "input_first_last_frame"),
|
||
}
|
||
case "last_frame":
|
||
return "模型能力未开启尾帧或首尾帧输入，已移除 last_frame。", capabilityPath(modelType, "input_first_last_frame"), map[string]any{
|
||
"input_last_frame": capabilityValue(context.modelCapability, modelType, "input_last_frame"),
|
||
"input_first_last_frame": capabilityValue(context.modelCapability, modelType, "input_first_last_frame"),
|
||
"max_images_for_last_frame": capabilityValue(context.modelCapability, modelType, "max_images_for_last_frame"),
|
||
"max_images_for_first_frame": capabilityValue(context.modelCapability, modelType, "max_images_for_first_frame"),
|
||
"max_images_for_middle_frame": capabilityValue(context.modelCapability, modelType, "max_images_for_middle_frame"),
|
||
}
|
||
case "reference_image":
|
||
return "模型能力未开启参考图输入，已移除 reference_image。", capabilityPath(modelType, "input_reference_generate_single"), map[string]any{
|
||
"input_reference_generate_single": capabilityValue(context.modelCapability, modelType, "input_reference_generate_single"),
|
||
"input_reference_generate_multiple": capabilityValue(context.modelCapability, modelType, "input_reference_generate_multiple"),
|
||
"max_images": capabilityValue(context.modelCapability, modelType, "max_images"),
|
||
}
|
||
default:
|
||
return "当前模型能力未开启图像输入，已移除 image_url。", capabilityPath(modelType, "input_first_frame"), map[string]any{
|
||
"input_first_frame": capabilityValue(context.modelCapability, modelType, "input_first_frame"),
|
||
"input_first_last_frame": capabilityValue(context.modelCapability, modelType, "input_first_last_frame"),
|
||
"input_reference_generate_single": capabilityValue(context.modelCapability, modelType, "input_reference_generate_single"),
|
||
"input_reference_generate_multiple": capabilityValue(context.modelCapability, modelType, "input_reference_generate_multiple"),
|
||
}
|
||
}
|
||
}
|
||
|
||
type inputAudioProcessor struct{}
|
||
|
||
func (inputAudioProcessor) Name() string { return "InputAudioProcessor" }
|
||
|
||
func (inputAudioProcessor) ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
if !isVideoModelType(modelType) {
|
||
return false
|
||
}
|
||
content := contentItems(params["content"])
|
||
for _, item := range content {
|
||
if isAudioContent(item) {
|
||
return true
|
||
}
|
||
}
|
||
return false
|
||
}
|
||
|
||
func (inputAudioProcessor) Process(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
content := contentItems(params["content"])
|
||
if len(content) == 0 {
|
||
return true
|
||
}
|
||
supportsInputAudio := false
|
||
if len(context.modelCapability) > 0 {
|
||
if isOmniVideoLike(context) {
|
||
supportsInputAudio = supportsOmniAudioReference(context)
|
||
} else if capability := capabilityForType(context.modelCapability, modelType); capability != nil {
|
||
supportsInputAudio = boolFromAny(capability["input_audio"])
|
||
}
|
||
}
|
||
if supportsInputAudio {
|
||
return true
|
||
}
|
||
next := make([]map[string]any, 0, len(content))
|
||
for index, item := range content {
|
||
if isAudioContent(item) {
|
||
path, value := audioInputCapabilityEvidence(context, modelType)
|
||
context.recordChange(
|
||
"InputAudioProcessor",
|
||
"remove",
|
||
fmt.Sprintf("content[%d]", index),
|
||
item,
|
||
nil,
|
||
"模型能力未开启输入音频，已移除 audio_url。",
|
||
path,
|
||
value,
|
||
)
|
||
continue
|
||
}
|
||
next = append(next, item)
|
||
}
|
||
params["content"] = mapsToAnySlice(next)
|
||
path, value := audioInputCapabilityEvidence(context, modelType)
|
||
deleteFieldsWithLog(params, context, "InputAudioProcessor", []string{"audio_url", "audioUrl", "reference_audio", "referenceAudio"}, "模型能力未开启输入音频，已移除音频参考快捷字段。", path, value)
|
||
return true
|
||
}
|
||
|
||
type durationProcessor struct{}
|
||
|
||
func (durationProcessor) Name() string { return "DurationProcessor" }
|
||
|
||
func (durationProcessor) ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
return isVideoModelType(modelType) && params["duration"] != nil
|
||
}
|
||
|
||
func (durationProcessor) Process(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
capability := capabilityForType(context.modelCapability, modelType)
|
||
if capability == nil {
|
||
return true
|
||
}
|
||
duration := floatFromAny(params["duration"])
|
||
if duration <= 0 {
|
||
return true
|
||
}
|
||
resolution := firstNonEmptyString(stringFromAny(params["resolution"]), context.resolution)
|
||
modeKey := videoModeKey(params)
|
||
if options := scopedNumberList(capability["duration_options"], resolution, modeKey); len(options) > 0 {
|
||
normalized := closestNumber(duration, options)
|
||
params["duration"] = normalized
|
||
syncDurationSeconds(params)
|
||
if normalized != duration {
|
||
context.recordChange(
|
||
"DurationProcessor",
|
||
"adjust",
|
||
"duration",
|
||
duration,
|
||
normalized,
|
||
"duration 不在模型固定时长选项内，已调整为最近的允许值。",
|
||
capabilityPath(modelType, "duration_options"),
|
||
capability["duration_options"],
|
||
)
|
||
}
|
||
return true
|
||
}
|
||
if minValue, maxValue, ok := scopedRange(capability["duration_range"], resolution, modeKey); ok {
|
||
step := durationStep(capability["duration_step"], resolution, modeKey)
|
||
normalized := normalizeDurationByRange(duration, minValue, maxValue, step)
|
||
params["duration"] = normalized
|
||
syncDurationSeconds(params)
|
||
if normalized != duration {
|
||
context.recordChange(
|
||
"DurationProcessor",
|
||
"adjust",
|
||
"duration",
|
||
duration,
|
||
normalized,
|
||
"duration 超出模型时长范围或步进配置，已按能力配置归一。",
|
||
capabilityPath(modelType, "duration_range"),
|
||
map[string]any{
|
||
"duration_range": capability["duration_range"],
|
||
"duration_step": capability["duration_step"],
|
||
},
|
||
)
|
||
}
|
||
}
|
||
return true
|
||
}
|
||
|
||
type audioProcessor struct{}
|
||
|
||
func (audioProcessor) Name() string { return "AudioProcessor" }
|
||
|
||
func (audioProcessor) ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
return isVideoModelType(modelType) && (params["audio"] != nil || params["output_audio"] != nil)
|
||
}
|
||
|
||
func (audioProcessor) Process(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
capability := capabilityForType(context.modelCapability, modelType)
|
||
if capability == nil || !boolFromAny(capability["output_audio"]) {
|
||
for _, key := range []string{"audio", "output_audio"} {
|
||
if before, ok := params[key]; ok {
|
||
delete(params, key)
|
||
context.recordChange(
|
||
"AudioProcessor",
|
||
"remove",
|
||
key,
|
||
before,
|
||
nil,
|
||
"模型能力未开启输出音频，已移除音频输出参数。",
|
||
capabilityPath(modelType, "output_audio"),
|
||
capabilityValue(context.modelCapability, modelType, "output_audio"),
|
||
)
|
||
}
|
||
}
|
||
}
|
||
return true
|
||
}
|
||
|
||
type imageCountProcessor struct{}
|
||
|
||
func (imageCountProcessor) Name() string { return "ImageCountProcessor" }
|
||
|
||
func (imageCountProcessor) ShouldProcess(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
return modelType == "image_generate" || modelType == "image_edit"
|
||
}
|
||
|
||
func (imageCountProcessor) Process(params map[string]any, modelType string, context *paramProcessContext) bool {
|
||
capability := capabilityForType(context.modelCapability, modelType)
|
||
if capability == nil || !boolFromAny(capability["output_multiple_images"]) {
|
||
return true
|
||
}
|
||
maxCount := int(math.Round(floatFromAny(capability["output_max_images_count"])))
|
||
if maxCount <= 0 {
|
||
return true
|
||
}
|
||
count := int(math.Round(floatFromAny(params["n"])))
|
||
if count <= 0 {
|
||
count = int(math.Round(floatFromAny(params["batch_size"])))
|
||
}
|
||
if count <= 0 {
|
||
count = 1
|
||
}
|
||
if count > maxCount {
|
||
before := count
|
||
count = maxCount
|
||
context.recordChange(
|
||
"ImageCountProcessor",
|
||
"adjust",
|
||
"n",
|
||
before,
|
||
count,
|
||
"请求图片数量超过模型输出上限，已按 output_max_images_count 截断。",
|
||
capabilityPath(modelType, "output_max_images_count"),
|
||
capability["output_max_images_count"],
|
||
)
|
||
}
|
||
params["n"] = count
|
||
return true
|
||
}
|
||
|
||
func ensureVideoContent(params map[string]any, context *paramProcessContext) {
|
||
if len(contentItems(params["content"])) > 0 {
|
||
return
|
||
}
|
||
content := make([]map[string]any, 0)
|
||
if prompt := firstNonEmptyString(stringFromAny(params["prompt"]), stringFromAny(params["input"])); prompt != "" {
|
||
content = append(content, map[string]any{"type": "text", "text": prompt})
|
||
}
|
||
appendURL := func(kind string, role string, url string) {
|
||
url = strings.TrimSpace(url)
|
||
if url == "" {
|
||
return
|
||
}
|
||
item := map[string]any{"type": kind, "role": role}
|
||
switch kind {
|
||
case "image_url":
|
||
item["image_url"] = map[string]any{"url": url}
|
||
case "video_url":
|
||
item["video_url"] = map[string]any{"url": url}
|
||
case "audio_url":
|
||
item["audio_url"] = map[string]any{"url": url}
|
||
}
|
||
content = append(content, item)
|
||
}
|
||
|
||
firstFrame := firstNonEmptyStringValue(params, "first_frame", "firstFrame")
|
||
appendURL("image_url", "first_frame", firstFrame)
|
||
appendURL("image_url", "last_frame", firstNonEmptyStringValue(params, "last_frame", "lastFrame"))
|
||
imageURLs := firstNonEmptyStringListFromAny(params["image"], params["images"], params["image_url"], params["imageUrl"], params["image_urls"], params["imageUrls"])
|
||
if firstFrame == "" && len(imageURLs) > 0 {
|
||
appendURL("image_url", "first_frame", imageURLs[0])
|
||
imageURLs = imageURLs[1:]
|
||
}
|
||
for _, url := range imageURLs {
|
||
appendURL("image_url", "reference_image", url)
|
||
}
|
||
for _, url := range firstNonEmptyStringListFromAny(params["reference_image"], params["referenceImage"]) {
|
||
appendURL("image_url", "reference_image", url)
|
||
}
|
||
for _, url := range firstNonEmptyStringListFromAny(params["video"], params["video_url"], params["videoUrl"], params["reference_video"], params["referenceVideo"]) {
|
||
appendURL("video_url", "reference_video", url)
|
||
}
|
||
for _, url := range firstNonEmptyStringListFromAny(params["audio_url"], params["audioUrl"], params["reference_audio"], params["referenceAudio"]) {
|
||
appendURL("audio_url", "reference_audio", url)
|
||
}
|
||
if len(content) > 0 {
|
||
params["content"] = mapsToAnySlice(content)
|
||
context.recordChange(
|
||
"ContentBuildProcessor",
|
||
"set",
|
||
"content",
|
||
nil,
|
||
params["content"],
|
||
"将 prompt/first_frame/reference_* 等快捷字段转换为 content 数组，后续处理器可按模型能力逐项过滤。",
|
||
"",
|
||
nil,
|
||
)
|
||
}
|
||
}
|
||
|
||
func effectiveModelCapability(candidate store.RuntimeModelCandidate) map[string]any {
|
||
base := cloneMap(candidate.Capabilities)
|
||
for key, value := range candidate.CapabilityOverride {
|
||
if baseChild, ok := base[key].(map[string]any); ok {
|
||
if overrideChild, ok := value.(map[string]any); ok {
|
||
base[key] = mergeMap(baseChild, overrideChild)
|
||
continue
|
||
}
|
||
}
|
||
base[key] = cloneAny(value)
|
||
}
|
||
return base
|
||
}
|
||
|
||
func filterUnsupportedOmniVideoContent(content []map[string]any, context *paramProcessContext) []map[string]any {
|
||
capability := omniVideoCapability(context)
|
||
maxVideos := math.Inf(1)
|
||
if capability != nil {
|
||
if value, ok := numericField(capability, "max_videos"); ok {
|
||
maxVideos = value
|
||
}
|
||
}
|
||
maxAudios := 0.0
|
||
if capability != nil {
|
||
if value, ok := numericField(capability, "max_audios"); ok {
|
||
maxAudios = value
|
||
} else if supportsOmniAudioReference(context) {
|
||
maxAudios = math.Inf(1)
|
||
}
|
||
}
|
||
|
||
videoCount := 0.0
|
||
audioCount := 0.0
|
||
out := make([]map[string]any, 0, len(content))
|
||
for index, item := range content {
|
||
if isVideoContent(item) {
|
||
if !supportsOmniVideoReference(item, capability) {
|
||
path, value := omniCapabilityEvidence(context, "supported_modes")
|
||
context.recordChange(
|
||
"ContentFilterProcessor",
|
||
"remove",
|
||
fmt.Sprintf("content[%d]", index),
|
||
item,
|
||
nil,
|
||
"视频参考类型不在 omni_video.supported_modes 允许范围内。",
|
||
path,
|
||
value,
|
||
)
|
||
continue
|
||
}
|
||
if videoCount >= maxVideos {
|
||
path, value := omniCapabilityEvidence(context, "max_videos")
|
||
context.recordChange(
|
||
"ContentFilterProcessor",
|
||
"remove",
|
||
fmt.Sprintf("content[%d]", index),
|
||
item,
|
||
nil,
|
||
"视频参考数量超过 omni_video.max_videos 限制。",
|
||
path,
|
||
value,
|
||
)
|
||
continue
|
||
}
|
||
videoCount++
|
||
out = append(out, item)
|
||
continue
|
||
}
|
||
if isAudioContent(item) {
|
||
if !supportsOmniAudioReference(context) {
|
||
path, value := omniCapabilityEvidence(context, "input_audio")
|
||
context.recordChange(
|
||
"ContentFilterProcessor",
|
||
"remove",
|
||
fmt.Sprintf("content[%d]", index),
|
||
item,
|
||
nil,
|
||
"模型能力不支持音频参考，已移除 audio_url。",
|
||
path,
|
||
mergeMetrics(map[string]any{"input_audio": value}, omniCapabilityBundle(context, "max_audios")),
|
||
)
|
||
continue
|
||
}
|
||
if audioCount >= maxAudios {
|
||
path, value := omniCapabilityEvidence(context, "max_audios")
|
||
context.recordChange(
|
||
"ContentFilterProcessor",
|
||
"remove",
|
||
fmt.Sprintf("content[%d]", index),
|
||
item,
|
||
nil,
|
||
"音频参考数量超过 omni_video.max_audios 限制。",
|
||
path,
|
||
value,
|
||
)
|
||
continue
|
||
}
|
||
audioCount++
|
||
out = append(out, item)
|
||
continue
|
||
}
|
||
out = append(out, item)
|
||
}
|
||
return out
|
||
}
|
||
|
||
func isOmniVideoLike(context *paramProcessContext) bool {
|
||
modelType := strings.TrimSpace(context.candidate.ModelType)
|
||
return modelType == "omni_video" ||
|
||
modelType == "omni" ||
|
||
context.modelCapability["omni_video"] != nil ||
|
||
context.modelCapability["omni"] != nil
|
||
}
|
||
|
||
func omniVideoCapability(context *paramProcessContext) map[string]any {
|
||
if capability := capabilityForType(context.modelCapability, "omni_video"); capability != nil {
|
||
return capability
|
||
}
|
||
return capabilityForType(context.modelCapability, "omni")
|
||
}
|
||
|
||
func supportsOmniAudioReference(context *paramProcessContext) bool {
|
||
capability := omniVideoCapability(context)
|
||
return capability != nil && (boolFromAny(capability["input_audio"]) || floatFromAny(capability["max_audios"]) > 0)
|
||
}
|
||
|
||
func supportsOmniVideoReference(item map[string]any, capability map[string]any) bool {
|
||
if capability == nil {
|
||
return true
|
||
}
|
||
if value, ok := numericField(capability, "max_videos"); ok && value == 0 {
|
||
return false
|
||
}
|
||
supportedModes := stringListFromAny(capability["supported_modes"])
|
||
supportsReference := containsString(supportedModes, "video_reference")
|
||
supportsEdit := containsString(supportedModes, "video_edit")
|
||
video, _ := item["video_url"].(map[string]any)
|
||
referType := stringFromAny(video["refer_type"])
|
||
isEditVideo := stringFromAny(item["role"]) == "video_base" || referType == "base"
|
||
isReferenceVideo := stringFromAny(item["role"]) == "video_feature" ||
|
||
stringFromAny(item["role"]) == "reference_video" ||
|
||
referType == "feature"
|
||
if isEditVideo {
|
||
return supportsEdit
|
||
}
|
||
if isReferenceVideo {
|
||
return supportsReference
|
||
}
|
||
return supportsReference || supportsEdit
|
||
}
|
||
|
||
func downgradeReferenceImageIfNeeded(params map[string]any, content []map[string]any, modelType string, context *paramProcessContext) {
|
||
if modelType != "image_to_video" && modelType != "video_generate" && modelType != "video_edit" && modelType != "omni_video" && modelType != "omni" {
|
||
return
|
||
}
|
||
if supportsReferenceImage(context.modelCapability, modelType) {
|
||
return
|
||
}
|
||
count := 0
|
||
for index, item := range content {
|
||
if stringFromAny(item["type"]) == "image_url" && stringFromAny(item["role"]) == "reference_image" {
|
||
before := cloneMap(item)
|
||
item["role"] = "first_frame"
|
||
context.recordChange(
|
||
"ContentFilterProcessor",
|
||
"adjust",
|
||
fmt.Sprintf("content[%d].role", index),
|
||
before,
|
||
item,
|
||
"模型不支持 reference_image，已降级为 first_frame。",
|
||
capabilityPath(modelType, "input_reference_generate_single"),
|
||
map[string]any{
|
||
"input_reference_generate_single": capabilityValue(context.modelCapability, modelType, "input_reference_generate_single"),
|
||
"input_reference_generate_multiple": capabilityValue(context.modelCapability, modelType, "input_reference_generate_multiple"),
|
||
"max_images": capabilityValue(context.modelCapability, modelType, "max_images"),
|
||
},
|
||
)
|
||
count++
|
||
}
|
||
}
|
||
if count > 0 {
|
||
appendParamWarning(params, "reference_image is unsupported by the selected model and was downgraded to first_frame")
|
||
}
|
||
}
|
||
|
||
func supportsReferenceImage(modelCapability map[string]any, modelType string) bool {
|
||
candidates := []map[string]any{}
|
||
if capability := capabilityForType(modelCapability, modelType); capability != nil {
|
||
candidates = append(candidates, capability)
|
||
}
|
||
if modelType != "image_to_video" {
|
||
if capability := capabilityForType(modelCapability, "image_to_video"); capability != nil {
|
||
candidates = append(candidates, capability)
|
||
}
|
||
}
|
||
if len(candidates) == 0 {
|
||
return true
|
||
}
|
||
for _, capability := range candidates {
|
||
_, hasSingle := capability["input_reference_generate_single"]
|
||
_, hasMultiple := capability["input_reference_generate_multiple"]
|
||
if hasSingle || hasMultiple {
|
||
if boolFromAny(capability["input_reference_generate_single"]) || boolFromAny(capability["input_reference_generate_multiple"]) {
|
||
return true
|
||
}
|
||
continue
|
||
}
|
||
if value, ok := numericField(capability, "max_images"); ok {
|
||
if value > 1 {
|
||
return true
|
||
}
|
||
continue
|
||
}
|
||
}
|
||
return false
|
||
}
|
||
|
||
func supportsFirstAndLastFrame(modelCapability map[string]any, modelType string) bool {
|
||
capability := capabilityForType(modelCapability, modelType)
|
||
if capability == nil {
|
||
return false
|
||
}
|
||
return boolFromAny(capability["input_first_last_frame"]) || floatFromAny(capability["max_images_for_last_frame"]) > 0
|
||
}
|
||
|
||
func validateAndAdjustAspectRatio(aspectRatio string, capability map[string]any, allowed []string) (string, bool) {
|
||
if !isMediaModelTypeWithAspectRatio(capability) {
|
||
return "", false
|
||
}
|
||
if ratioRange, ok := numberPair(capability["aspect_ratio_range"]); ok {
|
||
ratio, valid := aspectRatioNumber(aspectRatio)
|
||
if !valid || ratio < ratioRange[0] || ratio > ratioRange[1] {
|
||
return adjustAspectRatioToRange(aspectRatio, ratioRange[0], ratioRange[1], allowed), true
|
||
}
|
||
}
|
||
if allowed == nil {
|
||
return aspectRatio, true
|
||
}
|
||
if len(allowed) == 0 {
|
||
return "", false
|
||
}
|
||
if (aspectRatio == "adaptive" || aspectRatio == "keep_ratio") && !containsString(allowed, aspectRatio) {
|
||
return "", false
|
||
}
|
||
if containsString(allowed, aspectRatio) {
|
||
return aspectRatio, true
|
||
}
|
||
return allowed[0], true
|
||
}
|
||
|
||
func isMediaModelTypeWithAspectRatio(capability map[string]any) bool {
|
||
return capability != nil
|
||
}
|
||
|
||
func aspectRatioAllowed(value any, resolution string) []string {
|
||
switch typed := value.(type) {
|
||
case []any:
|
||
return stringListFromAny(typed)
|
||
case []string:
|
||
return typed
|
||
case map[string]any:
|
||
if resolution != "" {
|
||
if values := stringListFromAny(typed[resolution]); len(values) > 0 {
|
||
return values
|
||
}
|
||
}
|
||
return nil
|
||
default:
|
||
return nil
|
||
}
|
||
}
|
||
|
||
func scopedNumberList(value any, scopes ...string) []float64 {
|
||
switch typed := value.(type) {
|
||
case []any:
|
||
out := make([]float64, 0, len(typed))
|
||
for _, item := range typed {
|
||
if number := floatFromAny(item); number > 0 {
|
||
out = append(out, number)
|
||
}
|
||
}
|
||
return out
|
||
case []float64:
|
||
return typed
|
||
case []int:
|
||
out := make([]float64, 0, len(typed))
|
||
for _, item := range typed {
|
||
out = append(out, float64(item))
|
||
}
|
||
return out
|
||
case map[string]any:
|
||
for _, scope := range scopes {
|
||
if scope == "" {
|
||
continue
|
||
}
|
||
if values := scopedNumberList(typed[scope]); len(values) > 0 {
|
||
return values
|
||
}
|
||
}
|
||
for _, item := range typed {
|
||
if values := scopedNumberList(item); len(values) > 0 {
|
||
return values
|
||
}
|
||
}
|
||
}
|
||
return nil
|
||
}
|
||
|
||
func scopedRange(value any, scopes ...string) (float64, float64, bool) {
|
||
if pair, ok := numberPair(value); ok {
|
||
return pair[0], pair[1], true
|
||
}
|
||
if typed, ok := value.(map[string]any); ok {
|
||
for _, scope := range scopes {
|
||
if scope == "" {
|
||
continue
|
||
}
|
||
if minValue, maxValue, ok := scopedRange(typed[scope]); ok {
|
||
return minValue, maxValue, true
|
||
}
|
||
}
|
||
for _, item := range typed {
|
||
if minValue, maxValue, ok := scopedRange(item); ok {
|
||
return minValue, maxValue, true
|
||
}
|
||
}
|
||
}
|
||
return 0, 0, false
|
||
}
|
||
|
||
func durationStep(value any, scopes ...string) float64 {
|
||
if step := floatFromAny(value); step > 0 {
|
||
return step
|
||
}
|
||
if typed, ok := value.(map[string]any); ok {
|
||
for _, scope := range scopes {
|
||
if scope == "" {
|
||
continue
|
||
}
|
||
if step := durationStep(typed[scope]); step > 0 {
|
||
return step
|
||
}
|
||
}
|
||
for _, item := range typed {
|
||
if step := durationStep(item); step > 0 {
|
||
return step
|
||
}
|
||
}
|
||
}
|
||
return 0
|
||
}
|
||
|
||
func normalizeDurationByRange(target float64, minValue float64, maxValue float64, step float64) float64 {
|
||
clamped := math.Min(math.Max(target, minValue), maxValue)
|
||
if step <= 0 {
|
||
return clamped
|
||
}
|
||
snapped := math.Round((clamped-minValue)/step)*step + minValue
|
||
return math.Round(snapped*1_000_000) / 1_000_000
|
||
}
|
||
|
||
func closestNumber(target float64, values []float64) float64 {
|
||
if len(values) == 0 {
|
||
return target
|
||
}
|
||
closest := values[0]
|
||
minDiff := math.Abs(target - closest)
|
||
for _, value := range values[1:] {
|
||
diff := math.Abs(target - value)
|
||
if diff < minDiff {
|
||
minDiff = diff
|
||
closest = value
|
||
}
|
||
}
|
||
return closest
|
||
}
|
||
|
||
func videoModeKey(params map[string]any) string {
|
||
content := contentItems(params["content"])
|
||
hasFirstFrame := false
|
||
hasLastFrame := false
|
||
for _, item := range content {
|
||
switch stringFromAny(item["role"]) {
|
||
case "first_frame":
|
||
hasFirstFrame = true
|
||
case "last_frame":
|
||
hasLastFrame = true
|
||
}
|
||
}
|
||
switch {
|
||
case hasFirstFrame && hasLastFrame:
|
||
return "input_first_last_frame"
|
||
case hasFirstFrame:
|
||
return "input_first_frame"
|
||
case hasLastFrame:
|
||
return "input_last_frame"
|
||
default:
|
||
return ""
|
||
}
|
||
}
|
||
|
||
func syncDurationSeconds(params map[string]any) {
|
||
if params["duration_seconds"] != nil {
|
||
params["duration_seconds"] = params["duration"]
|
||
}
|
||
}
|
||
|
||
func syncVideoConvenienceFields(params map[string]any, content []map[string]any, context *paramProcessContext) {
|
||
hasVideo := false
|
||
hasAudio := false
|
||
for _, item := range content {
|
||
hasVideo = hasVideo || isVideoContent(item)
|
||
hasAudio = hasAudio || isAudioContent(item)
|
||
}
|
||
if !hasVideo {
|
||
path, value := omniCapabilityEvidence(context, "supported_modes")
|
||
deleteFieldsWithLog(params, context, "ContentFilterProcessor", []string{"video", "video_url", "videoUrl", "reference_video", "referenceVideo"}, "对应视频 content 已被模型能力过滤，移除视频参考快捷字段。", path, value)
|
||
}
|
||
if !hasAudio {
|
||
path, value := omniCapabilityEvidence(context, "input_audio")
|
||
deleteFieldsWithLog(params, context, "ContentFilterProcessor", []string{"audio_url", "audioUrl", "reference_audio", "referenceAudio"}, "对应音频 content 已被模型能力过滤，移除音频参考快捷字段。", path, mergeMetrics(map[string]any{"input_audio": value}, omniCapabilityBundle(context, "max_audios")))
|
||
}
|
||
}
|
||
|
||
func deleteFieldsWithLog(params map[string]any, context *paramProcessContext, processor string, keys []string, reason string, capabilityPath string, capabilityValue any) {
|
||
for _, key := range keys {
|
||
if before, ok := params[key]; ok {
|
||
delete(params, key)
|
||
context.recordChange(processor, "remove", key, before, nil, reason, capabilityPath, capabilityValue)
|
||
}
|
||
}
|
||
}
|
||
|
||
func appendParamWarning(params map[string]any, warning string) {
|
||
warnings, _ := params["_param_warnings"].([]any)
|
||
for _, item := range warnings {
|
||
if stringFromAny(item) == warning {
|
||
return
|
||
}
|
||
}
|
||
params["_param_warnings"] = append(warnings, warning)
|
||
}
|
||
|
||
func filterContent(content []map[string]any, keep func(map[string]any) bool) []map[string]any {
|
||
out := make([]map[string]any, 0, len(content))
|
||
for _, item := range content {
|
||
if keep(item) {
|
||
out = append(out, item)
|
||
}
|
||
}
|
||
return out
|
||
}
|
||
|
||
func contentItems(value any) []map[string]any {
|
||
switch typed := value.(type) {
|
||
case []any:
|
||
out := make([]map[string]any, 0, len(typed))
|
||
for _, item := range typed {
|
||
if object, ok := item.(map[string]any); ok {
|
||
out = append(out, cloneMap(object))
|
||
}
|
||
}
|
||
return out
|
||
case []map[string]any:
|
||
out := make([]map[string]any, 0, len(typed))
|
||
for _, item := range typed {
|
||
out = append(out, cloneMap(item))
|
||
}
|
||
return out
|
||
default:
|
||
return nil
|
||
}
|
||
}
|
||
|
||
func mapsToAnySlice(values []map[string]any) []any {
|
||
out := make([]any, 0, len(values))
|
||
for _, value := range values {
|
||
out = append(out, value)
|
||
}
|
||
return out
|
||
}
|
||
|
||
func isImageContent(item map[string]any) bool {
|
||
return stringFromAny(item["type"]) == "image_url" || item["image_url"] != nil
|
||
}
|
||
|
||
func isVideoContent(item map[string]any) bool {
|
||
return stringFromAny(item["type"]) == "video_url" || item["video_url"] != nil
|
||
}
|
||
|
||
func isAudioContent(item map[string]any) bool {
|
||
return stringFromAny(item["type"]) == "audio_url" || item["audio_url"] != nil
|
||
}
|
||
|
||
func capabilityForType(capabilities map[string]any, modelType string) map[string]any {
|
||
if capabilities == nil {
|
||
return nil
|
||
}
|
||
if typed, ok := capabilities[modelType].(map[string]any); ok {
|
||
return typed
|
||
}
|
||
return nil
|
||
}
|
||
|
||
func capabilityPath(modelType string, key string) string {
|
||
modelType = strings.TrimSpace(modelType)
|
||
if modelType == "" {
|
||
return ""
|
||
}
|
||
if strings.TrimSpace(key) == "" {
|
||
return "capabilities." + modelType
|
||
}
|
||
return "capabilities." + modelType + "." + key
|
||
}
|
||
|
||
func capabilityValue(capabilities map[string]any, modelType string, key string) any {
|
||
capability := capabilityForType(capabilities, modelType)
|
||
if capability == nil {
|
||
return nil
|
||
}
|
||
return cloneAny(capability[key])
|
||
}
|
||
|
||
func capabilityEvidence(capabilities map[string]any, modelType string, key string) (string, any) {
|
||
return capabilityPath(modelType, key), capabilityValue(capabilities, modelType, key)
|
||
}
|
||
|
||
func audioInputCapabilityEvidence(context *paramProcessContext, modelType string) (string, any) {
|
||
if isOmniVideoLike(context) {
|
||
path, value := omniCapabilityEvidence(context, "input_audio")
|
||
return path, mergeMetrics(map[string]any{"input_audio": value}, omniCapabilityBundle(context, "max_audios"))
|
||
}
|
||
return capabilityEvidence(context.modelCapability, modelType, "input_audio")
|
||
}
|
||
|
||
func omniCapabilityType(context *paramProcessContext) string {
|
||
if context != nil && capabilityForType(context.modelCapability, "omni_video") != nil {
|
||
return "omni_video"
|
||
}
|
||
if context != nil && capabilityForType(context.modelCapability, "omni") != nil {
|
||
return "omni"
|
||
}
|
||
return "omni_video"
|
||
}
|
||
|
||
func omniCapabilityEvidence(context *paramProcessContext, key string) (string, any) {
|
||
modelType := omniCapabilityType(context)
|
||
var capabilities map[string]any
|
||
if context != nil {
|
||
capabilities = context.modelCapability
|
||
}
|
||
return capabilityPath(modelType, key), capabilityValue(capabilities, modelType, key)
|
||
}
|
||
|
||
func omniCapabilityBundle(context *paramProcessContext, keys ...string) map[string]any {
|
||
modelType := omniCapabilityType(context)
|
||
var capabilities map[string]any
|
||
if context != nil {
|
||
capabilities = context.modelCapability
|
||
}
|
||
out := map[string]any{}
|
||
for _, key := range keys {
|
||
out[key] = capabilityValue(capabilities, modelType, key)
|
||
}
|
||
return out
|
||
}
|
||
|
||
func numericField(values map[string]any, key string) (float64, bool) {
|
||
if values == nil {
|
||
return 0, false
|
||
}
|
||
if _, ok := values[key]; !ok {
|
||
return 0, false
|
||
}
|
||
return floatFromAny(values[key]), true
|
||
}
|
||
|
||
func boolFromAny(value any) bool {
|
||
typed, _ := value.(bool)
|
||
return typed
|
||
}
|
||
|
||
func firstNonEmptyStringValue(values map[string]any, keys ...string) string {
|
||
for _, key := range keys {
|
||
if value := stringFromAny(values[key]); value != "" {
|
||
return value
|
||
}
|
||
}
|
||
return ""
|
||
}
|
||
|
||
func firstNonEmptyStringListFromAny(values ...any) []string {
|
||
for _, value := range values {
|
||
items := stringListFromAny(value)
|
||
if len(items) > 0 {
|
||
return items
|
||
}
|
||
}
|
||
return nil
|
||
}
|
||
|
||
func stringListFromAny(value any) []string {
|
||
switch typed := value.(type) {
|
||
case []string:
|
||
out := make([]string, 0, len(typed))
|
||
for _, item := range typed {
|
||
if text := strings.TrimSpace(item); text != "" {
|
||
out = append(out, text)
|
||
}
|
||
}
|
||
return out
|
||
case []any:
|
||
out := make([]string, 0, len(typed))
|
||
for _, item := range typed {
|
||
if text := stringFromAny(item); text != "" {
|
||
out = append(out, text)
|
||
}
|
||
}
|
||
return out
|
||
case string:
|
||
if strings.TrimSpace(typed) == "" {
|
||
return nil
|
||
}
|
||
return []string{strings.TrimSpace(typed)}
|
||
default:
|
||
return nil
|
||
}
|
||
}
|
||
|
||
func containsString(values []string, target string) bool {
|
||
for _, value := range values {
|
||
if value == target {
|
||
return true
|
||
}
|
||
}
|
||
return false
|
||
}
|
||
|
||
func appendUniqueString(values *[]string, value string) {
|
||
value = strings.TrimSpace(value)
|
||
if value == "" {
|
||
return
|
||
}
|
||
for _, existing := range *values {
|
||
if existing == value {
|
||
return
|
||
}
|
||
}
|
||
*values = append(*values, value)
|
||
}
|
||
|
||
func numberPair(value any) ([2]float64, bool) {
|
||
switch typed := value.(type) {
|
||
case []any:
|
||
if len(typed) < 2 {
|
||
return [2]float64{}, false
|
||
}
|
||
return [2]float64{floatFromAny(typed[0]), floatFromAny(typed[1])}, true
|
||
case []float64:
|
||
if len(typed) < 2 {
|
||
return [2]float64{}, false
|
||
}
|
||
return [2]float64{typed[0], typed[1]}, true
|
||
case []int:
|
||
if len(typed) < 2 {
|
||
return [2]float64{}, false
|
||
}
|
||
return [2]float64{float64(typed[0]), float64(typed[1])}, true
|
||
default:
|
||
return [2]float64{}, false
|
||
}
|
||
}
|
||
|
||
func validAspectRatio(value string) bool {
|
||
if value == "adaptive" || value == "keep_ratio" {
|
||
return true
|
||
}
|
||
_, ok := aspectRatioNumber(value)
|
||
return ok
|
||
}
|
||
|
||
func aspectRatioNumber(value string) (float64, bool) {
|
||
parts := strings.Split(value, ":")
|
||
if len(parts) != 2 {
|
||
return 0, false
|
||
}
|
||
width := parsePositiveFloat(parts[0])
|
||
height := parsePositiveFloat(parts[1])
|
||
if width <= 0 || height <= 0 {
|
||
return 0, false
|
||
}
|
||
return width / height, true
|
||
}
|
||
|
||
func adjustAspectRatioToRange(value string, minValue float64, maxValue float64, allowed []string) string {
|
||
current, ok := aspectRatioNumber(value)
|
||
if !ok {
|
||
if len(allowed) > 0 {
|
||
return allowed[0]
|
||
}
|
||
return "1:1"
|
||
}
|
||
if len(allowed) > 0 {
|
||
closest := ""
|
||
minDiff := math.Inf(1)
|
||
for _, candidate := range allowed {
|
||
ratio, ok := aspectRatioNumber(candidate)
|
||
if !ok || ratio < minValue || ratio > maxValue {
|
||
continue
|
||
}
|
||
diff := math.Abs(ratio - current)
|
||
if diff < minDiff {
|
||
minDiff = diff
|
||
closest = candidate
|
||
}
|
||
}
|
||
if closest != "" {
|
||
return closest
|
||
}
|
||
}
|
||
if current < minValue {
|
||
return ratioString(minValue)
|
||
}
|
||
return ratioString(maxValue)
|
||
}
|
||
|
||
func ratioString(value float64) string {
|
||
if value <= 0 {
|
||
return "1:1"
|
||
}
|
||
return strings.TrimRight(strings.TrimRight(strconv.FormatFloat(value, 'f', 6, 64), "0"), ".") + ":1"
|
||
}
|
||
|
||
func parsePositiveFloat(value string) float64 {
|
||
for _, r := range strings.TrimSpace(value) {
|
||
if r < '0' || r > '9' {
|
||
if r != '.' {
|
||
return 0
|
||
}
|
||
}
|
||
}
|
||
out, _ := strconv.ParseFloat(strings.TrimSpace(value), 64)
|
||
return out
|
||
}
|
||
|
||
func isEmptyParamString(value string) bool {
|
||
normalized := strings.ToLower(strings.TrimSpace(value))
|
||
return normalized == "null" || normalized == "undefined"
|
||
}
|
||
|
||
func isImageResolution(modelType string, value string) bool {
|
||
return (modelType == "image_generate" || modelType == "image_edit") && containsString([]string{"1K", "2K", "4K", "8K"}, value)
|
||
}
|
||
|
||
func isVideoResolution(modelType string, value string) bool {
|
||
return isVideoModelType(modelType) && containsString([]string{"480p", "720p", "1080p", "1440p", "2160p"}, value)
|
||
}
|
||
|
||
func isVideoModelType(modelType string) bool {
|
||
return modelType == "video_generate" || modelType == "text_to_video" || modelType == "image_to_video" || modelType == "video_edit" || modelType == "omni_video" || modelType == "omni"
|
||
}
|
||
|
||
func cloneMap(values map[string]any) map[string]any {
|
||
out := map[string]any{}
|
||
for key, value := range values {
|
||
out[key] = cloneAny(value)
|
||
}
|
||
return out
|
||
}
|
||
|
||
func cloneAny(value any) any {
|
||
switch typed := value.(type) {
|
||
case map[string]any:
|
||
return cloneMap(typed)
|
||
case []any:
|
||
out := make([]any, 0, len(typed))
|
||
for _, item := range typed {
|
||
out = append(out, cloneAny(item))
|
||
}
|
||
return out
|
||
case []map[string]any:
|
||
out := make([]any, 0, len(typed))
|
||
for _, item := range typed {
|
||
out = append(out, cloneMap(item))
|
||
}
|
||
return out
|
||
default:
|
||
return value
|
||
}
|
||
}
|