The recent PR that added resize_cond_for_context_window methods to
model classes used inline 'import comfy.context_windows' in each
method body. This moves that import to the top-level import section,
replacing 4 duplicate inline imports with a single top-level one.
* Add slice_cond and per-model context window cond resizing
* Fix cond_value.size() call in context window cond resizing
* Expose additional advanced inputs for ContextWindowsManualNode
Necessary for WanAnimate context windows workflow, which needs cond_retain_index_list = 0 to work properly with its reference input.
---------
* chore(api-nodes): mark seedream-3-0-t2i and seedance-1-0-lite models as deprecated
* fix(api-nodes): fixed old regression in the ByteDanceImageReference node
---------
Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>
* sd: soft_empty_cache on tiler fallback
This doesnt cost a lot and creates the expected VRAM reduction in
resource monitors when you fallback to tiler.
* wan: vae: Don't recursion in local fns (move run_up)
Moved Decoder3d’s recursive run_up out of forward into a class
method to avoid nested closure self-reference cycles. This avoids
cyclic garbage that delays garbage of tensors which in turn delays
VRAM release before tiled fallback.
* ltx: vae: Don't recursion in local fns (move run_up)
Mov the recursive run_up out of forward into a class
method to avoid nested closure self-reference cycles. This avoids
cyclic garbage that delays garbage of tensors which in turn delays
VRAM release before tiled fallback.
* ltx: vae: add cache state to downsample block
* ltx: vae: Add time stride awareness to causal_conv_3d
* ltx: vae: Automate truncation for encoder
Other VAEs just truncate without error. Do the same.
* sd/ltx: Make chunked_io a flag in its own right
Taking this bi-direcitonal, so make it a for-purpose named flag.
* ltx: vae: implement chunked encoder + CPU IO chunking
People are doing things with big frame counts in LTX including V2V
flows. Implement the time-chunked encoder to keep the VRAM down, with
the converse of the new CPU pre-allocation technique, where the chunks
are brought from the CPU JIT.
* ltx: vae-encode: round chunk sizes more strictly
Only powers of 2 and multiple of 8 are valid due to cache slicing.
* ltx: vae: add cache state to downsample block
* ltx: vae: Add time stride awareness to causal_conv_3d
* ltx: vae: Automate truncation for encoder
Other VAEs just truncate without error. Do the same.
* sd/ltx: Make chunked_io a flag in its own right
Taking this bi-direcitonal, so make it a for-purpose named flag.
* ltx: vae: implement chunked encoder + CPU IO chunking
People are doing things with big frame counts in LTX including V2V
flows. Implement the time-chunked encoder to keep the VRAM down, with
the converse of the new CPU pre-allocation technique, where the chunks
are brought from the CPU JIT.
* ltx: vae-encode: round chunk sizes more strictly
Only powers of 2 and multiple of 8 are valid due to cache slicing.
On Apple Silicon, `vram_state` is set to `VRAMState.SHARED` because
CPU and GPU share unified memory. However, `text_encoder_device()`
only checked for `HIGH_VRAM` and `NORMAL_VRAM`, causing all text
encoders to fall back to CPU on MPS devices.
Adding `VRAMState.SHARED` to the condition allows non-quantized text
encoders (e.g. bf16 Gemma 3 12B) to run on the MPS GPU, providing
significant speedup for text encoding and prompt generation.
Note: quantized models (fp4/fp8) that use float8_e4m3fn internally
will still fall back to CPU via the `supports_cast()` check in
`CLIP.__init__()`, since MPS does not support fp8 dtypes.
* wan: vae: encoder: Add feature cache layer that corks singles
If a downsample only gives you a single frame, save it to the feature
cache and return nothing to the top level. This increases the
efficiency of cacheability, but also prepares support for going two
by two rather than four by four on the frames.
* wan: remove all concatentation with the feature cache
The loopers are now responsible for ensuring that non-final frames are
processes at least two-by-two, elimiating the need for this cat case.
* wan: vae: recurse and chunk for 2+2 frames on decode
Avoid having to clone off slices of 4 frame chunks and reduce the size
of the big 6 frame convolutions down to 4. Save the VRAMs.
* wan: encode frames 2x2.
Reduce VRAM usage greatly by encoding frames 2 at a time rather than
4.
* wan: vae: remove cloning
The loopers now control the chunking such there is noever more than 2
frames, so just cache these slices directly and avoid the clone
allocations completely.
* wan: vae: free consumer caller tensors on recursion
* wan: vae: restyle a little to match LTX
* ltx: vae: scale the chunk size with the users VRAM
Scale this linearly down for users with low VRAM.
* ltx: vae: free non-chunking recursive intermediates
* ltx: vae: cleanup some intermediates
The conv layer can be the VRAM peak and it does a torch.cat. So cleanup
the pieces of the cat. Also clear our the cache ASAP as each layer detect
its end as this VAE surges in VRAM at the end due to the ended padding
increasing the size of the final frame convolutions off-the-books to
the chunker. So if all the earlier layers free up their cache it can
offset that surge.
Its a fragmentation nightmare, and the chance of it having to recache the
pyt allocator is very high, but you wont OOM.
Mark the weight_dtype parameter in UNETLoader (Load Diffusion Model) as
an advanced input to reduce UI complexity for new users. The parameter
is now hidden behind an expandable Advanced section, matching the
pattern used for other advanced inputs like device, tile_size, and
overlap.
Amp-Thread-ID: https://ampcode.com/threads/T-019cbaf1-d3c0-718e-a325-318baba86dec
Write to a temp file in the same directory then os.replace() onto the
target path. If the process crashes mid-write, the original file is
left intact instead of being truncated to zero bytes.
Fixes#11298
Add a GitHub Actions workflow and shell script that scan all commits
in a pull request for Co-authored-by trailers from known AI coding
agents (Claude, Cursor, Copilot, Codex, Aider, Devin, Gemini, Jules,
Windsurf, Cline, Amazon Q, Continue, OpenCode, etc.).
The check fails with clear instructions on how to remove the trailers
via interactive rebase.
* feat(assets): align local API with cloud spec
Unify response models, add missing fields, and align input schemas with
the cloud OpenAPI spec at cloud.comfy.org/openapi.
- Replace AssetSummary/AssetDetail/AssetUpdated with single Asset model
- Add is_immutable, metadata (system_metadata), prompt_id fields
- Support mime_type and preview_id in update endpoint
- Make CreateFromHashBody.name optional, add mime_type, require >=1 tag
- Add id/mime_type/preview_id to upload, relax tags to optional
- Rename total_tags → tags in tag add/remove responses
- Add GET /api/assets/tags/refine histogram endpoint
- Add DB migration for system_metadata and prompt_id columns
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Fix review issues: tags validation, size nullability, type annotation, hash mismatch check, and add tag histogram tests
- Remove contradictory min_length=1 from CreateFromHashBody.tags default
- Restore size field to int|None=None for proper null semantics
- Add Union type annotation to _build_asset_response result param
- Add hash mismatch validation on idempotent upload path (409 HASH_MISMATCH)
- Add unit tests for list_tag_histogram service function
Amp-Thread-ID: https://ampcode.com/threads/T-019cd993-f43c-704e-b3d7-6cfc3d4d4a80
Co-authored-by: Amp <amp@ampcode.com>
* Add preview_url to /assets API response using /api/view endpoint
For input and output assets, generate a preview_url pointing to the
existing /api/view endpoint using the asset's filename and tag-derived
type (input/output). Handles subdirectories via subfolder param and
URL-encodes filenames with spaces, unicode, and special characters.
This aligns the OSS backend response with the frontend AssetCard
expectation for thumbnail rendering.
Amp-Thread-ID: https://ampcode.com/threads/T-019cda3f-5c2c-751a-a906-ac6c9153ac5c
Co-authored-by: Amp <amp@ampcode.com>
* chore: remove unused imports from asset_reference queries
Amp-Thread-ID: https://ampcode.com/threads/T-019cda7d-cb21-77b4-a51b-b965af60208c
Co-authored-by: Amp <amp@ampcode.com>
* feat: resolve blake3 hashes in /view endpoint via asset database
Amp-Thread-ID: https://ampcode.com/threads/T-019cda7d-cb21-77b4-a51b-b965af60208c
Co-authored-by: Amp <amp@ampcode.com>
* Register uploaded images in asset database when --enable-assets is set
Add register_file_in_place() service function to ingest module for
registering already-saved files without moving them. Call it from the
/upload/image endpoint to return asset metadata in the response.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce023-3384-7560-bacf-de40b0de0dd2
Co-authored-by: Amp <amp@ampcode.com>
* Exclude None fields from asset API JSON responses
Add exclude_none=True to model_dump() calls across asset routes to
keep response payloads clean by omitting unset optional fields.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce023-3384-7560-bacf-de40b0de0dd2
Co-authored-by: Amp <amp@ampcode.com>
* Add comment explaining why /view resolves blake3 hashes
Amp-Thread-ID: https://ampcode.com/threads/T-019ce023-3384-7560-bacf-de40b0de0dd2
Co-authored-by: Amp <amp@ampcode.com>
* Move blake3 hash resolution to asset_management service
Extract resolve_hash_to_path() into asset_management.py and remove
_resolve_blake3_to_path from server.py. Also revert loopback origin
check to original logic.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce023-3384-7560-bacf-de40b0de0dd2
Co-authored-by: Amp <amp@ampcode.com>
* Require at least one tag in UploadAssetSpec
Enforce non-empty tags at the Pydantic validation layer so uploads
with no tags are rejected with a 400 before reaching ingest. Adds
test_upload_empty_tags_rejected to cover this case.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce377-8bde-7048-bc28-a9df063409f9
Co-authored-by: Amp <amp@ampcode.com>
* Add owner_id check to resolve_hash_to_path
Filter asset references by owner visibility so the /view endpoint
only resolves hashes for assets the requesting user can access.
Adds table-driven tests for owner visibility cases.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce377-8bde-7048-bc28-a9df063409f9
Co-authored-by: Amp <amp@ampcode.com>
* Make ReferenceData.created_at and updated_at required
Remove None defaults and type: ignore comments. Move fields before
optional fields to satisfy dataclass ordering.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce377-8bde-7048-bc28-a9df063409f9
Co-authored-by: Amp <amp@ampcode.com>
* Fix double commit in create_from_hash
Move mime_type update into _register_existing_asset so it shares a
single transaction with reference creation. Log a warning when the
hash is not found instead of silently returning None.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce377-8bde-7048-bc28-a9df063409f9
Co-authored-by: Amp <amp@ampcode.com>
* Add exclude_none=True to create/upload responses
Align with get/update/list endpoints for consistent JSON output.
Amp-Thread-ID: https://ampcode.com/threads/T-019ce377-8bde-7048-bc28-a9df063409f9
Co-authored-by: Amp <amp@ampcode.com>
* Change preview_id to reference asset by reference ID, not content ID
Clients receive preview_id in API responses but could not dereference it
through public routes (which use reference IDs). Now preview_id is a
self-referential FK to asset_references.id so the value is directly
usable in the public API.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Filter soft-deleted and missing refs from visibility queries
list_references_by_asset_id and list_tags_with_usage were not filtering
out deleted_at/is_missing refs, allowing /view?filename=blake3:... to
serve files through hidden references and inflating tag usage counts.
Add list_all_file_paths_by_asset_id for orphan cleanup which
intentionally needs unfiltered access.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Pass preview_id and mime_type through all asset creation fast paths
The duplicate-content upload path and hash-based creation paths were
silently dropping preview_id and mime_type. This wires both fields
through _register_existing_asset, create_from_hash, and all route
call sites so behavior is consistent regardless of whether the asset
content already exists.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Remove unimplemented client-provided ID from upload API
The `id` field on UploadAssetSpec was advertised for idempotent creation
but never actually honored when creating new references. Remove it
rather than implementing the feature.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Make asset mime_type immutable after first ingest
Prevents cross-tenant metadata mutation when multiple references share
the same content-addressed Asset row. mime_type can now only be set when
NULL (first ingest); subsequent attempts to change it are silently ignored.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Use resolved content_type from asset lookup in /view endpoint
The /view endpoint was discarding the content_type computed by
resolve_hash_to_path() and re-guessing from the filename, which
produced wrong results for extensionless files or mismatched extensions.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Merge system+user metadata into filter projection
Extract rebuild_metadata_projection() to build AssetReferenceMeta rows
from {**system_metadata, **user_metadata}, so system-generated metadata
is queryable via metadata_filter and user keys override system keys.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Standardize tag ordering to alphabetical across all endpoints
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Derive subfolder tags from path in register_file_in_place
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Reject client-provided id, fix preview URLs, rename tags→total_tags
- Reject 'id' field in multipart upload with 400 UNSUPPORTED_FIELD
instead of silently ignoring it
- Build preview URL from the preview asset's own metadata rather than
the parent asset's
- Rename 'tags' to 'total_tags' in TagsAdd/TagsRemove response schemas
for clarity
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: SQLite migration 0003 FK drop fails on file-backed DBs (MB-2)
Add naming_convention to Base.metadata so Alembic batch-mode reflection
can match unnamed FK constraints created by migration 0002. Pass
naming_convention and render_as_batch=True through env.py online config.
Add migration roundtrip tests (upgrade/downgrade/cycle from baseline).
Amp-Thread-ID: https://ampcode.com/threads/T-019ce466-1683-7471-b6e1-bb078223cda0
Co-authored-by: Amp <amp@ampcode.com>
* Fix missing tag count for is_missing references and update test for total_tags field
- Allow is_missing=True references to be counted in list_tags_with_usage
when the tag is 'missing', so the missing tag count reflects all
references that have been tagged as missing
- Add update_is_missing_by_asset_id query helper for bulk updates by asset
- Update test_add_and_remove_tags to use 'total_tags' matching the API schema
Amp-Thread-ID: https://ampcode.com/threads/T-019ce482-05e7-7324-a1b0-a56a929cc7ef
Co-authored-by: Amp <amp@ampcode.com>
* Remove unused imports in scanner.py
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Rename prompt_id to job_id on asset_references
Rename the column in the DB model, migration, and service schemas.
The API response emits both job_id and prompt_id (deprecated alias)
for backward compatibility with the cloud API.
Amp-Thread-ID: https://ampcode.com/threads/T-019cef41-60b0-752a-aa3c-ed7f20fda2f7
Co-authored-by: Amp <amp@ampcode.com>
* Add index on asset_references.preview_id for FK cascade performance
Amp-Thread-ID: https://ampcode.com/threads/T-019cef45-a4d2-7548-86d2-d46bcd3db419
Co-authored-by: Amp <amp@ampcode.com>
* Add clarifying comments for Asset/AssetReference naming and preview_id
Amp-Thread-ID: https://ampcode.com/threads/T-019cef49-f94e-7348-bf23-9a19ebf65e0d
Co-authored-by: Amp <amp@ampcode.com>
* Disallow all-null meta rows: add CHECK constraint, skip null values on write
- convert_metadata_to_rows returns [] for None values instead of an all-null row
- Remove dead None branch from _scalar_to_row
- Simplify null filter in common.py to just check for row absence
- Add CHECK constraint ck_asset_reference_meta_has_value to model and migration 0003
Amp-Thread-ID: https://ampcode.com/threads/T-019cef4e-5240-7749-bb25-1f17fcf9c09c
Co-authored-by: Amp <amp@ampcode.com>
* Remove dead None guards on result.asset in upload handler
register_file_in_place guarantees a non-None asset, so the
'if result.asset else None' checks were unreachable.
Amp-Thread-ID: https://ampcode.com/threads/T-019cef5b-4cf8-723c-8a98-8fb8f333c133
Co-authored-by: Amp <amp@ampcode.com>
* Remove mime_type from asset update API
Clients can no longer modify mime_type after asset creation via the
PUT /api/assets/{id} endpoint. This reduces the risk of mime_type
spoofing. The internal update_asset_hash_and_mime function remains
available for server-side use (e.g., enrichment).
Amp-Thread-ID: https://ampcode.com/threads/T-019cef5d-8d61-75cc-a1c6-2841ac395648
Co-authored-by: Amp <amp@ampcode.com>
* Fix migration constraint naming double-prefix and NULL in mixed metadata lists
- Use fully-rendered constraint names in migration 0003 to avoid the
naming convention doubling the ck_ prefix on batch operations.
- Add table_args to downgrade so SQLite batch mode can find the CHECK
constraint (not exposed by SQLite reflection).
- Fix model CheckConstraint name to use bare 'has_value' (convention
auto-prefixes).
- Skip None items when converting metadata lists to rows, preventing
all-NULL rows that violate the has_value check constraint.
Amp-Thread-ID: https://ampcode.com/threads/T-019cef87-94f9-7172-a6af-c6282290ce4f
Co-authored-by: Amp <amp@ampcode.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Amp <amp@ampcode.com>
* feat: add essentials_category to nodes and blueprints for Essentials tab
Add ESSENTIALS_CATEGORY or essentials_category to 12 node classes and all
36 blueprint JSONs. Update SubgraphEntry TypedDict and subgraph_manager to
extract and pass through the field.
Fixes COM-15221
Amp-Thread-ID: https://ampcode.com/threads/T-019c83de-f7ab-7779-a451-0ba5940b56a9
* fix: import NotRequired from typing_extensions for Python 3.10 compat
* refactor: keep only node class ESSENTIALS_CATEGORY, remove blueprint/subgraph changes
Frontend will own blueprint categorization separately.
* fix: remove essentials_category from CreateVideo (not in spec)
---------
Co-authored-by: guill <jacob.e.segal@gmail.com>
If a subclass BYO _load_from_state_dict and doesnt call the super() the
needed default init of these weights is missed and can lead to problems
for uninitialized weights.
This is an experimental WIP option that might not work in your workflow but
should lower memory usage if it does.
Currently only the VAE and the load image node will output in fp16 when
this option is turned on.
After a frontend update (e.g. nightly build), browsers could load
outdated cached index.html and JS/CSS chunks, causing dynamically
imported modules to fail with MIME type errors and vite:preloadError.
Hard refresh (Ctrl+Shift+R) was insufficient to fix the issue because
Cache-Control: no-cache still allows the browser to cache and
revalidate via ETags. aiohttp's FileResponse auto-generates ETags
based on file mtime+size, which may not change after pip reinstall,
so the browser gets 304 Not Modified and serves stale content.
Clearing ALL site data in DevTools did fix it, confirming the HTTP
cache was the root cause.
The fix changes:
- index.html: no-cache -> no-store, must-revalidate
- JS/CSS/JSON entry points: no-cache -> no-store
no-store instructs browsers to never cache these responses, ensuring
every page load fetches the current index.html with correct chunk
references. This is a small tradeoff (~5KB re-download per page load)
for guaranteed correctness after updates.
