ComfyUI — Image & Video Generation#

Use to generate images (Flux schnell) and videos (Wan2.1 T2V/I2V) on the local RTX 5080 GPU.

Environment#

• ComfyUI: ~/ComfyUI (systemd user service: comfyui.service)
• Python venv: ~/comfyui-venv
• API: http://127.0.0.1:8188
• Output: ~/ComfyUI/output/

Script#

text

{baseDir}/scripts/generate.py <mode> [options]

Mode: image — Text-to-Image (Flux schnell)#

bash

{baseDir}/scripts/generate.py image \
  --prompt "A cat on the moon" \
  --output /tmp/output.png

OptionDefaultDescription--prompt(required)Text prompt--negative""Negative prompt--width1024Image width--height1024Image height--steps4Sampling steps (schnell optimized)--seedrandomReproducible seed--outputComfyUI output dirCopy output here--modelflux1-schnell.safetensorsUNET filename--weight-dtypefp8_e4m3fnWeight quantization--wait120Max wait seconds

Recommended Flux schnell params: steps=4, cfg=1.0, sampler=euler, scheduler=simple

Mode: t2v — Text-to-Video (Wan2.1 T2V-1.3B)#

bash

{baseDir}/scripts/generate.py t2v \
  --prompt "A red sports car driving on a mountain road at sunset" \
  --length 49 \
  --output /tmp/video_frames/

OptionDefaultDescription--prompt(required)Text prompt--negative""Negative prompt--width832Frame width--height480Frame height--length49Number of frames (≈3s at 16fps)--steps20Sampling steps--seedrandomReproducible seed--outputComfyUI output dirCopy frames here--wait300Max wait seconds

Recommended Wan2.1 T2V params: steps=20, cfg=5.0, sampler=uni_pc_bh2, scheduler=simple

Mode: i2v — Image-to-Video (Wan2.1 I2V using T2V-1.3B)#

bash

{baseDir}/scripts/generate.py i2v \
  --prompt "gentle wave motion, water flowing" \
  --image /path/to/input.png \
  --output /tmp/video_frames/

OptionDefaultDescription--prompt(required)Motion description--image(required)Path to input image--length49Number of frames--steps20Sampling steps--seedrandomReproducible seed--outputComfyUI output dirCopy frames here--wait300Max wait seconds

Server Management#

bash

# Start (systemd user service)
systemctl --user start comfyui.service

# Check status
systemctl --user status comfyui.service

# Check API
curl -s http://127.0.0.1:8188/system_stats | python3 -m json.tool

# Manual start (if systemd not available)
cd ~/ComfyUI && LD_LIBRARY_PATH=~/comfyui-venv/lib/python3.12/site-packages/nvidia/cuda_runtime/lib:$LD_LIBRARY_PATH ~/comfyui-venv/bin/python main.py --listen 127.0.0.1 --port 8188

Installed Models#

Image (Flux)#

FileLocationSizeflux1-schnell.safetensorsmodels/unet/23.8GBae.safetensorsmodels/vae/335MBclip_l.safetensorsmodels/clip/250MBt5xxl_fp16.safetensorsmodels/clip/9.8GB

Video (Wan2.1)#

FileLocationSizewan2.1_t2v_1.3B_bf16.safetensorsmodels/diffusion_models/5.3GBwan2.1_vae.pthmodels/vae/485MBumt5_xxl_fp8_e4m3fn_scaled.safetensorsmodels/text_encoders/6.1GBopen_clip_xlm_roberta_large_vit_huge_14.pthmodels/clip/4.5GB (for I2V)

Workflow#

1. Check ComfyUI status ( curl http://127.0.0.1:8188/system_stats).
2. Start if needed ( systemctl --user start comfyui.service).
3. Call generate.py with appropriate mode and options.
4. Return output image/frames to user; offer xdg-open to view.
5. For video: frames are individual PNGs; optionally combine into MP4 with imageio.

Troubleshooting#

• libcudart.so not found: set LD_LIBRARY_PATH with nvidia/cuda_runtime/lib.
• OOM on 16GB VRAM: reduce resolution or use lower --length for video.
• Video generation slow: T2V-1.3B 49 frames ≈ 2-3 minutes on RTX 5080.
• Server won't start: pkill -f "main.py" for stale processes.
• All models downloaded from ModelScope (domestic) — HuggingFace inaccessible.