COMFYUI LOCAL VIDEO GENERATION SETUP GUIDE (NILAY)

---------------------------------------
1. SYSTEM REQUIREMENTS
---------------------------------------
GPU: NVIDIA RTX 4070 (8GB VRAM)
Drivers: Installed (nvidia-smi working)
Python: 3.10+
OS: Windows

---------------------------------------
2. INSTALLATION STEPS
---------------------------------------

Step 1: Clone ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI

Step 2: Create Virtual Environment
python -m venv comfy-env
comfy-env\Scripts\activate

Step 3: Install Dependencies
pip install -r requirements.txt

Step 4: Install CUDA-enabled PyTorch
pip uninstall torch torchvision torchaudio -y
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Step 5: Verify GPU
python
import torch
print(torch.cuda.is_available())
print(torch.cuda.get_device_name(0))

---------------------------------------
3. RUN COMFYUI
---------------------------------------
python main.py

Open in browser:
http://127.0.0.1:8188

---------------------------------------
4. MODELS SETUP
---------------------------------------

IMPORTANT:
Do NOT use SVD model for text prompts.

Use:

A. Text-to-Image Model (REQUIRED)
Download:
v1-5-pruned-emaonly.safetensors
Place in:
ComfyUI/models/checkpoints/

B. Video Model (OPTIONAL)
svd_xt.safetensors
(Used ONLY for image-to-video)

---------------------------------------
5. BASIC WORKFLOW (TEXT → IMAGE)
---------------------------------------

Nodes:
- Load Checkpoint (SD model)
- CLIP Text Encode (positive)
- CLIP Text Encode (negative)
- Empty Latent Image
- KSampler
- VAE Decode
- Save Image

Connections:

Checkpoint.CLIP → CLIP Encode (both +ve & -ve)
Checkpoint.MODEL → KSampler.model
Checkpoint.VAE → VAE Decode.vae

Positive → KSampler.positive
Negative → KSampler.negative
Latent → KSampler.latent_image

KSampler → VAE Decode
VAE Decode → Save

---------------------------------------
6. IMPORTANT SETTINGS
---------------------------------------

Resolution: 512 x 512
Batch size (frames): 16
Steps: 20
CFG: 7.5
Sampler: euler
Scheduler: normal
Denoise: 1.0

---------------------------------------
7. COMMON ERRORS & FIXES
---------------------------------------

Error: Torch not compiled with CUDA
→ Install CUDA version of PyTorch

Error: steps = NaN
→ Delete and re-add KSampler

Error: clip input is invalid
→ Wrong model (SVD used instead of SD)

Error: CUDA out of memory
→ Reduce resolution or batch size

---------------------------------------
8. VIDEO GENERATION PIPELINE (CORRECT)
---------------------------------------

Text → Image (SD model)
Image → Video (SVD / AnimateDiff)

NOT:
Text → Video directly (will fail)

---------------------------------------
9. OUTPUT LOCATION
---------------------------------------

Generated files:
ComfyUI/output/

---------------------------------------
END