comfyui_propainter_nodes: Uncensored Long Video Generation with Sam 3, Wan 2.2 & Painter 12V
Get clean, on-brand videos from images or short clips. comfyui_propainter_nodes let you extend, inpaint, and fix frames with scene-level consistency. Pair them with Sam 3 for masks, Wan 2.2 for generation, and Painter 12V for stable tracking. You get a reliable pipeline for dynamic image-to-video conversion, plus easy 4K upscaling.
What are comfyui_propainter_nodes and why use them?
They are ComfyUI nodes for video inpainting, extension, and object cleanup with temporal stability. You remove or replace objects, extend scenes, and keep details consistent across frames.
Use them when you need frame-locked edits, logo swaps, sky fixes, or safe, uncensored long-form output.
How do you install them with ComfyUI Manager?
Install from the Manager so dependencies resolve cleanly.
Open ComfyUI, go to Manager, search Propainter, click Install, then Update All and Restart. If you deploy on Runpod, do the same inside the container and persist the custom_nodes folder.
What is a proven long-video workflow with Sam 3, Wan 2.2, and Painter 12V?
You combine robust masking, a strong video generator, and a tracker that holds your edits over time.
Run Sam 3 to segment targets, feed masks to Propainter for clean fills, drive motion with Wan 2.2, and keep masks steady with Painter 12V.
Workflow, high-level
- Input image or short clip, define prompt and duration
- Sam 3, auto masks for subject and background regions
- ProPainter, remove or replace elements with a temporal fill
- Wan 2.2, generate motion and new frames from the edited base
- Painter 12V, refine mask tracks for long shots
- Output frames, then encode to video with constant fps
How do you build a dynamic image-to-video pipeline in ComfyUI?
Start from a simple graph, then add tracking and inpainting where needed.
Load image, set prompt and negative prompt, generate first frames, add propainter branch for regions that need fixes, then combine and write video at 24–30 fps.
Graph blocks to include
- Load Image or Load Video
- Sam 3 Mask Generator
- ProPainter Inpaint Video
- Wan 2.2 Video Generator
- Painter 12V Tracker
- Video Combine, Write Video
How do you upscale AI video to 4K cleanly?
Upscale after all edits are done to avoid re-painting artifacts.
Export frames at native size, run a quality upscaler, then re-encode. Keep sharpening light and bitrate high.
4K steps
- Export PNG frames from ComfyUI
- Run a 2× upscaler pass, then a light detail pass
- Re-assemble at 3840 × 2160, 24–30 fps, high bitrate
Where does Huanyuan 1.5 fit in this stack?
Use it for efficient motion or when VRAM is tight.
Generate base motion with Huanyuan 1.5, then apply propainter for object fixes. If you need richer detail, hand off to Wan 2.2 for a second pass.
What do TSV & 12V mean in practice?
They refer to tracker variants that stabilize masks across time.
Use 12V tracking on long shots with slow camera moves. Use faster TSV-style passes for quick edits or short transitions.
How do you keep the pipeline stable on Runpod?
Pin versions and keep the Manager tidy.
Launch a GPU pod, install ComfyUI Manager, add comfyui_propainter_nodes, Sam 3, Wan 2.2, and Painter 12V, then snapshot the image. Persist models and custom_nodes to a volume.
Can Qwen edit prompts and shot notes on the fly?
Yes. Use a small Qwen model to refine prompts and negative prompts by beat.
Have it suggest short lens words, light words, and one motion cue per shot to reduce drift.
Step-by-step: long video creation from a single image
You move from still to stable motion in minutes.
- Load the image and write a short prompt with one motion cue
- Generate a 4–6 second base clip with Wan 2.2
- Use Sam 3 to segment objects to remove or protect
- Apply propainter to fill or swap regions across frames
- Track with Painter 12V to keep masks stable
- Export frames, upscale to 4K, then encode to MP4
Common pitfalls and quick fixes
Small errors compound over many frames, so fix inputs early.
- Overlong prompts cause drift; keep them short
- Too many motion cues introduce warp; use one
- Heavy sharpening creates halos; keep it light
- Cropped captions, use safe zones before final encode
Model comparison for long-video work
| Model | Best use | Strength | VRAM feel | Notes |
| Sam 3 | Mask generation | Fast, accurate segmentation | Low | Use as a pre-mask step |
| ProPainter nodes | Inpaint across frames | Temporal consistency | Mid | Core of cleanup and extension |
| Wan 2.2 | Video generation | Rich detail, stable motion | Mid-High | Great second-pass motion |
| Huanyuan 1.5 | Efficient base motion | Low VRAM, fast drafts | Low-Mid | Good first pass before upscales |
| Painter 12V | Mask tracking | Long-shot stability | Mid | Keeps edits locked over time |
Settings that keep results clean
A few small choices raise quality fast.
- 24 or 30 fps, fixed from the start
- One lens word, one light word, one motion cue
- Negative prompt to avoid blur and double edges
- Upscale once at the end, not mid-pipe
Quick recipe: product demo without logos
You remove marks, extend the background, and keep the motion natural.
- Sam 3 masks logos and reflections
- ProPainter fills brand-safe regions
- Wan 2.2 adds slow turntable motion
- Painter 12V keeps edges stable
- Upscale to 4K, add captions, export
FAQs
What are comfyui_propainter_nodes used for?
Frame-consistent video inpainting. You remove objects, extend scenes, and keep textures aligned over time.
How do I install them with ComfyUI Manager?
Open Manager, search “propainter,” click Install, then Update All and Restart. Repeat this on Runpod and persist custom_nodes.
Can I build long videos from a single image?
Yes. Generate motion with Wan 2.2 or Huanyuan 1.5, then fix regions with propainter and track masks with Painter 12V.
How do I upscale to 4K without artefacts?
Export PNG frames, run a 2× upscaler, apply light detail only, then re-encode at 3840 × 2160 with a high bitrate.
When should I use Sam 3 in the pipeline?
At the start. Segment targets and protections first, so the painter and the video model keep edges clean.
Does this work with ComfyUI Manager on a shared GPU?
Yes. Install nodes via Manager, keep models on a persistent volume, and lock versions for repeatability.
