Starter workflows
On first run WorkflowRemix copies six example workflows into your ComfyUI workflows folder and switches them on, so Remix has something to run.
Expect to swap the models
Each workflow names the checkpoints, LoRAs and custom nodes its author used. Open one in ComfyUI first to see which custom nodes you are missing. In WorkflowRemix, a model your ComfyUI does not have shows as an amber pill, ready to be replaced with one you do have. Models are never downloaded for you, and each carries a licence of its own that you must follow.
Text to image
Krea2
Text to image · the simplest one
A prompt in, a picture out, in eight steps. Thirteen nodes: a model, its text encoder and VAE, a LoRA stack and a sampler. The one to start with, and the easiest to read in ComfyUI.
- File
APP Krea2.json- Form
- Prompt, seed, steps, size, model, LoRAs
- Needs
- A Krea 2 turbo diffusion model with its text encoder and VAE
T2I
Text to image · anime, with detail passes
A full illustration pipeline on an SDXL-family checkpoint: a fast first pass, a high-resolution second pass, then detailers that find and redraw faces and hands, and an upscale at the end. Good for seeing how Remix reduces a graph of about 150 nodes to a short form.
- File
APP T2I.json- Form
- Prompt and negative prompt, seed, size, checkpoint, LoRAs, the detail and upscale passes
- Needs
- An SDXL or Illustrious checkpoint and VAE, face and hand detector models, an upscale model, and the rgthree, Impact Pack, Efficiency Nodes and Easy-Use node packs among others
ANIMA T2I
Text to image · presets
The same shape as T2I, built for the Anima model family, with groups of ready-made style and quality presets: anime, semi-real, realism. Tick several presets and each one runs as a pass of its own.
- File
APP ANIMA T2I.json- Form
- Prompt, seed, size, model, LoRAs, style and quality presets
- Needs
- An Anima diffusion model with its text encoder and VAE, detector and upscale models, and the same node packs as T2I
From an image
REAL
Image to image · photoreal
Takes a picture and remakes it as a photograph. ControlNet holds the composition (edges, depth, pose), IPAdapter and FaceID carry a likeness across, and a refiner, face detailers and an upscale finish it. With Remix's Match Input Image ticked, each run is sized to the picture going in.
- File
APP REAL.json- Form
- Input image, prompt, seed, checkpoint, LoRAs, the control and refine stages
- Needs
- SDXL checkpoints, a ControlNet union model, IPAdapter and FaceID models, detector and upscale models, and the node packs that go with them
Krea2 Identity Edit
Image edit · by instruction
You give it a picture and say what to change: “make the jacket red”, “put her in a winter street”. Its prompt is an instruction and not a description, so WorkflowRemix leaves the workflow's own prompt in place when you switch to it, where it would otherwise carry the picture's prompt over. An optional second image is a person to place into the scene; that part stays off until you pick one.
- File
APP Krea2 Identity Edit.json- Form
- Image to edit, optional second reference, instruction, seed, steps
- Needs
- The Krea 2 turbo model, its text encoder and VAE, and the Krea 2 identity edit LoRA
Image to video
VIDEO CLIP SFW
Image to video · WAN 2.2
A still in, a short clip out. WAN 2.2 uses a pair of models, one for the high-noise steps and one for the low-noise steps, each with a LoRA stack of its own. Remix treats a high and low LoRA pair as one: add it to one stack and its partner is added to the other.
- File
APP VIDEO CLIP SFW.json- Form
- Input image, prompt, seed, the high and low models and their LoRAs
- Needs
- WAN 2.2 image-to-video models (high and low noise), their text encoder, CLIP vision model and VAE, and the GGUF, rgthree and KJNodes node packs
Using your own
Nothing about these six is special. Any workflow you save in ComfyUI appears on the Workflows page, where one tick offers it in Remix. The form is worked out from the graph, so there is nothing to convert.