Three Scene Generator

Here's a node that needs an honest intro: the Three Scene Generator does not generate three scenes. Read the name, wonder, then read this. It takes three scene descriptions, all the usual model and sampler inputs, runs a full encode pass on each scene - and then throws the results away and hands you three solid-color images. Red, green, and blue. The README calls them "placeholder images for testing/demo," and the source code literally contains the comment #REplace this shit. It's a stub.

What it's actually for

The useful reading is "workflow dry-run." Since the real pipeline (encode → KSample → VAE decode) is slow and VRAM-heavy - the author runs this against FLUX on a 96GB GPU - you can drop this node in to verify your storyboard layout, spacing, and labels work before committing to a full generation. That's a legitimately handy thing to have. The trap is mistaking it for a working generator and building your real workflow on it.

How it works (the disappointing part)

The source encodes all three scenes into conditioning internally, prints a debug line saying it "outputs conditioning for use with sampling nodes" - which it also doesn't do - then creates three constant-color tensors sized to your width/height and returns them as image_1, image_2, image_3. Every parameter on the node (steps, cfg, seed, sampler_name, scheduler, negative_prompt) is accepted and ignored. Seed does nothing. Nothing is sampled.

The inputs, since you'll still see them

• scene_1, scene_2, scene_3 - the descriptions, usually from SceneParser.
• model, clip, vae - wired like any sampling chain; required even though the sampler never runs.
• width / height - this is the one pair that genuinely matters, because it sets the placeholder size.
• Everything else - steps, cfg, seed, sampler, scheduler, and the optional negative_prompt - is cosmetic here.

Outputs are image_1, image_2, image_3 as IMAGE tensors. Feed them to StoryboardCompositor and you'll get a color-blocked mockup you can check before burning real GPU time.

Installing

Same pack as everything else in FairyTaler:

cd ComfyUI/custom_nodes
git clone https://github.com/IIEleven11/ComfyUI-FairyTaler.git

Restart ComfyUI, find it under FairyTaler/Storyboard. No requirements.txt, no downloads, no keys.

Where people get burned

The number one "why are my images solid colors?" post this node generates is really a "this node never generated anything" moment. If you want real images, skip it entirely and use the README's basic workflow instead: SceneParser → three CLIP encodes → three KSamplers → three VAE decodes → StoryboardCompositor. If you want to check that your storyboard composition looks right before spending a minute per panel on FLUX, this is exactly the tool. Just don't expect it to do the thing its name promises.