🎨 S42 CutFlow Style Transfer
Eight cinematic color grades, one dropdown, zero model downloads
- clip
- clip
- frame_count
- info
"Style Transfer" is a misleading name - there's no neural style transfer here, no VGG, no model. What S42CF_StyleTransfer actually is: eight opinionated cinematic color-grade presets you can apply to any clip with a dropdown and a strength slider. That's a good thing. It means it's instant, deterministic, runs on CPU, and you get exactly the look the preset names promise. It's the fastest way in S42 CutFlow to take raw AI output and make it read as "filmed."
The presets
- cinematic_teal_orange (default) - the Hollywood blockbuster look: warm highlights, cool shadows. If you apply only one preset ever, this is it; it's the complementary-color scheme that dominates modern blockbusters.
- vintage - faded warm tones with lifted blacks. That washed-out retro-photo feel.
- noir - high-contrast, desaturated. Black-and-white-adjacent moodiness.
- polaroid - warm shadows, cool highlights. The instant-film look.
- cyberpunk - neon-shifted, high saturation. Synthwave city at night.
- pastel - soft, desaturated pastels. Gentle and clean.
- bleach_bypass - silver-retention film: high contrast, low saturation, the Saving Private Ryan desaturated-war look.
- cross_process - shifted color channels, the retro film-lab-accident look.
The inputs
Two, and that's the beauty of it:
- style - the eight presets above.
- strength (float, 0–1, default 1) - blends between your original (0) and the fully styled result (1). This is the control that saves you from over-cooking everything: 0.3 gives a subtle enhancement, 1.0 goes full grade. For AI footage, starting at 0.5–0.7 is usually the sweet spot - full strength on already-saturated LTX output can clip.
Outputs: clip, frame_count, and info. Nothing else - you feed a clip, you get a graded clip of the same length.
Installing it
Standard S42 CutFlow install:
cd ComfyUI/custom_nodes/
git clone https://github.com/GeekyGhost/S42-CutFlow.git
pip install -r S42-CutFlow/requirements.txt # Windows portable: python_embeded\python.exe -m pip
Or ComfyUI Manager → search S42 CutFlow → install → restart. It's under S42 CutFlow/Filters.
Gotchas
The name does people dirty, so let me be blunt: if you came here expecting neural style transfer (content image + style image → merged painting), this isn't that. It's a deterministic color science, which is honestly more useful for video work - neural style transfer on video flickers horribly frame to frame, and a deterministic grade doesn't. The trap that does bite: applying a grade before your temporal cleanup. Grade first, then a film-grain or denoise pass on top, and the grain breaks up the graded look's banding; grade after, and you amplify noise into the shadows. Order matters. Also, strength 1.0 on every preset isn't "best" - the presets are opinionated by design, so trust the strength slider to back off. And since this is one of the pack's "no model" nodes, it plays fine on a 6GB card in a long chain; throw it at the end of an LTX post-process and it behaves exactly like a traditional LUT would.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | IMAGE | Video clip to stylize. | |
| style | COMBO | cinematic_teal_orange | Color style preset: 'cinematic_teal_orange' = Hollywood blockbuster look. 'vintage' = faded warm tones, lifted blacks. 'noir' = high-contrast desaturated. 'polaroid' = warm shadows, cool highlights. 'cyberpunk' = neon-shifted, high saturation. 'pastel' = soft, desaturated pastels. 'bleach_bypass' = silver-retention film look. 'cross_process' = shifted color channels, retro. |
| strength | FLOAT | 1.000–1 | Blend between original (0) and fully styled (1). |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| clip | IMAGE | — |
| frame_count | INT | — |
| info | STRING | — |