FairyTaler Storyboard

Straight answer first: the README marks this node BROKEN - it's literally struck through in the node list. FairyTaler Storyboard is the pack's attempt at an all-in-one: feed it the raw Ollama text, get your three parsed scenes and a composed storyboard image from one node. It's the kind of convenience every beginner wants. The author just doesn't trust it, and you shouldn't either.

What it tries to do

It's SceneParser and StoryboardCompositor fused into one class. You give it ollama_text, it regex-splits it into three scenes using the same Scene 1: / Scene 2: / Scene 3: markers, applies scene constants (manual or auto-extracted), and then - depending on whether you fed it images - either pastes image_1/2/3 into a storyboard or renders an 800×600 placeholder board that prints "Connect images to create visual storyboard" and a preview of each scene's text. It doesn't generate images itself at all; the three panels have to arrive from somewhere else.

The inputs and outputs

Required: ollama_text, plus the compositor controls layout, spacing, background_color, add_labels, and debug.

Optional: image_1, image_2, image_3 (if you provide all three, you get a visual storyboard instead of the text placeholder), plus scene_constants, constants_position, and constants_format for the consistency injection.

Outputs: scene_1, scene_2, scene_3 (STRING), storyboard (IMAGE), and extracted_constants (STRING).

Installing

It's part of the ComfyUI-FairyTaler pack:

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

Restart ComfyUI and it appears under FairyTaler/Storyboard. No requirements.txt (torch, numpy, and Pillow are all ComfyUI stock), no model downloads, no keys.

Why you should reach for the pieces instead

The all-in-one collapses the graph but also collapses your control. Because it never generates panels, the "storyboard" half only works if you've built the generation pipeline elsewhere anyway - at which point the all-in-one saved you almost nothing and removed your ability to inspect the intermediate scenes before they hit the sampler. The README's own tips say the same thing: use the individual nodes for maximum control, keep the all-in-one for quick testing only. If your storyboard output comes back as a white box full of text, that's not a bug - you simply didn't connect images, and that's the node doing its (limited) thing.

So treat FairyTaler Storyboard as a preview of the pack's intended UX rather than a production node. Build the classic chain instead - SceneParser → CLIP encodes → KSamplers → VAE decode → StoryboardCompositor - and you'll have every stage inspectable, every failure debuggable, and none of the author's own "this is broken" asterisks hanging over your workflow.