InvictusRedmond: A Generalist Stable Cascade Finetune
InvictusRedmond is a generalist Stable Cascade Stage C finetune for realistic, artistic, car, and character images. See how to run it in ComfyUI.
On this page
InvictusRedmond: A Generalist Stable Cascade Finetune
InvictusRedmond is a Stable Cascade checkpoint by ArtificialGuyBR released in March 2024 as a generalist Stage C finetune. Unlike most Stable Cascade community models, it is not built around a single subject or style - it targets realistic and artistic images, cars, people, and a wide variety of themes.
- Stage C (prior) checkpoint for the Stable Cascade pipeline, not an SDXL model.
- v1.1 "less training" is the recommended version; it beats v1.0 on quality while staying easier to control.
- No trigger words: prompt it like a standard generalist checkpoint in ComfyUI.
What InvictusRedmond Does
The model is a fine-tune of Stable Cascade, the three-stage architecture from Stability AI:
- Stage C - the text-conditioned prior that produces compressed latents
- Stage B - the diffusion decoder
- Stage A - the image decoder/VAE
InvictusRedmond replaces Stage C. The author describes the result as "high capacity to generate realistic, artistic images, cars, people, and a wide variety of themes" (Civitai model page). Its tag set - anime, realistic, female, woman, game character, photorealistic, generalist - points to a flexible model for portraits and characters across styles.
Why v1.1 - Less Training - Is the Recommended Version
Two versions are published. The author's own notes on v1.1: it shows better results than the base model and v1.0, but is "a little more similar to the original version of Stable Cascade," and the model is "extremely sensitive to higher training runs even with low LR" (Civitai version notes). The lighter training run is the sweet spot - more training drifted the model or degraded output, so less was more here.
How to Use InvictusRedmond
For Stable Cascade in ComfyUI, the Stage C prior model goes in the models/unet folder, and the workflow chains Stage C into the Stage B decoder and the previewer VAE. InvictusRedmond ships as a 6.69 GB BF16 SafeTensor file (InvictusRedmond-StageC-Updated.safetensors) - a single Stage C checkpoint with no trigger words.
The author confirms no special prompt entry is needed: use standard descriptive prompts as you would with any generalist model. Pair it with a Stable Cascade text encoder for the prior stage, then decode through Stage B and the previewer.
Common errors and fixes
Error | Cause | Fix | Source
|---|---|---|---
Wrong output or crash in SDXL tools | InvictusRedmond is a Stable Cascade Stage C checkpoint, not SDXL | Use a Stable Cascade workflow (Stage C + Stage B + previewer/VAE) in ComfyUI | Civitai page (base model: Stable Cascade) v1.0 results look worse than expected | v1.0 was overtrained - training runs are extremely sensitive even at low LR | Use v1.1 - Less training | Civitai version notes No text encoder / VAE in checkpoint | Stage C is only the prior part of the pipeline | Load the Stable Cascade text encoder and previewer VAE separately | SeaArt / Civitai model page
FAQ
Is InvictusRedmond an SDXL model? No. It is a Stage C finetune for Stable Cascade, so it needs the Cascade Stage B decoder and previewer VAE (Civitai, base model: Stable Cascade).
Does it need a trigger word? No trigger words are published; prompt it directly like a generalist checkpoint (Civitai trained words: none).
Which version should I download? v1.1 - Less training (March 1, 2024) - the author recommends it over v1.0 and the base model.
Is there a recommended sampler or resolution? None published for this checkpoint; follow standard Stable Cascade settings in your workflow.
Sources
- InvictusRedmond on Civitai - model page, version notes, base model, file details
- InvictusRedmond on SeaArt - description mirror and version history
- Stable Cascade on GitHub - three-stage pipeline reference