Files
comfyui-workflows/README.md
T

113 lines
4.0 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# ComfyUI Workflows
Model-organized workflow library for local ComfyUI use.
## Layout
- `models/flux/` - FLUX family workflows and notes
- `models/krea2/` - Krea 2 workflows and notes
- `models/ltx/` - LTX workflows and notes
- `models/flux2klein/` - FLUX.2-klein workflows and notes
- `models/ideogram/` - Ideogram workflows and text-placement notes
- `models/minimax/` - MiniMax workflows and notes
- `custom/` - personal or in-house working workflows kept separate from imports and masters
- `optimized/` - maintained category masters built from collected workflows
- `prompts/` - reusable generic prompts, matching workflow baselines, and generated results
- `models/_incoming/` - temporary holding area when a workflow is still being classified
- `templates/workflow-template/` - starter folder for new workflow imports
## How workflows are stored
Each workflow gets its own folder so it can carry:
- `workflow.json` - the actual ComfyUI workflow file
- `README.md` - what it does, source link, model/version notes, required custom nodes, setup quirks, and a record of meaningful workflow parameters and changes
- optional supporting files such as screenshots, example prompts, or helper notes
Suggested path pattern:
`models/<model-family>/<workflow-slug>/`
Example:
`models/ltx/first-frame-character-lock/`
Optimized workflows live under:
`optimized/<category>/<family>-master/`
Personal or project-specific working workflows live under:
`custom/<owner>/<workflow-slug>/`
## Conventions
- Organize primarily by model family.
- Prefer one workflow per folder.
- Keep source links in each workflow README.
- Note whether the workflow is `tested`, `imported-only`, or `needs-fixes`.
- Record required custom nodes so the workflow is easier to revive later.
- Record the important generation and edit settings whenever a workflow is tuned or promoted: positive prompt, negative prompt, CFG, steps, sampler, scheduler, seed, denoise, resolution, model, VAE, LoRAs, ControlNets, and node-level changes that materially affect output.
## Generic prompts
Reusable prompt experiments live under [`prompts/`](prompts/). Each prompt has
its own folder containing the prompt text, workflow baseline, README, and a
`results/` folder for generated images and metadata. Prompt iterations stay
separate from the maintained optimized workflow masters.
## Initial model families
- `flux`
- `krea2`
- `ltx`
- `flux2klein`
- `ideogram`
- `minimax`
More model families can be added later as needed.
## Optimized layer
The `optimized/` section is the maintained layer for cleaned-up "best current" workflows by category.
These are not raw imports. They are intended to become:
- better organized
- better documented
- stripped of junk or dead branches
- updated when stronger ideas appear in newly collected workflows
Expected categories:
- `text-to-image`
- `image-edit`
- `image-to-video`
- `text-to-video`
- `upscale-restoration`
Each optimized workflow should explain:
- what source workflows informed it
- what changes were made
- why those changes were chosen
- what still needs real-world testing
- what run settings or node parameters define the current behavior
## Apples-to-apples model comparisons
Within each category, model matrices must use the category's canonical
fixtures. T2I comparisons use the shared prompt in `optimized/T2I/prompt.txt`;
I2I comparisons use `optimized/I2I/character_reference.png` and the shared
I2I prompt; MultipleRef2I comparisons use
`optimized/MultipleRef2I/person_reference.png`,
`optimized/MultipleRef2I/scene_reference.png`, and the shared multi-reference
prompt. A model-specific matrix may vary settings such as resolution, steps,
CFG, sampler, scheduler, or seed, but it must not silently change the prompt
or reference inputs when claiming a cross-model comparison.
Category comparison resolutions are also fixed: T2I uses `1088×1920`, I2I
uses a `1080×1920` output target, and MultipleRef2I uses `1920×1080`. A
separate resolution-scaling study may vary resolution, but it must be clearly
labelled as a scaling study rather than a cross-model comparison.