HoochMix MOSS v1.0

Anime • Semi-Realistic • Realistic | MOSS-Image Merge

HoochMix MOSS is an experimental checkpoint merge in BF16 precision.

The goal is to create a versatile image generation model capable of producing detailed anime illustrations, cinematic fantasy artwork, semi-realistic characters, and realistic-looking images without being restricted to a single visual style.

Visual Style & Capabilities

HoochMix MOSS supports several visual directions:

• Anime: expressive characters, detailed outfits, fantasy illustrations, cinematic scenes and vibrant environments.

• Semi-Realistic: a blend of anime aesthetics and realistic anatomy, lighting, skin rendering and materials.

• Realistic: photographic-looking portraits, fashion, glamour, natural environments and realistic lighting.

Style flexibility is one of its interesting characteristics.

Without explicit style tags, the model may naturally produce anime, semi-realistic or realistic-looking results depending on the prompt, seed and composition.

For example, realistic-looking beach portraits have been generated using simple prompts without explicit photorealistic or realistic tags.

HoochMix MOSS is not marketed as a dedicated photorealistic checkpoint, and results may vary.

Recommended Generation Settings

Steps 25

CFG 2.5

Sampler Euler

Scheduler Simple

Model Sampling Shift 5.0

Skip Layer GuidanceOptional SLG Layers 22, 24, 26

SLG Start / End0.35 / 0.70

Prompt EnhancerOptional — OFF recommended for baseline testing

Optional Hi-Res Refiner

The accompanying workflow includes an optional latent refinement stage.

Current experimental settings:

Latent Upscale 1.25×

Upscale Method Bicubic

Refiner Steps 10 (or 8)

Refiner CFG 1.0

Refiner Sampler Euler

Refiner Scheduler Simple

Hi-Res Denoise 0.50

The Refiner is optional. Some images may develop doubled outlines, ghosting or altered details during refinement. The 1.25× upscale setting has shown promising results but is still being evaluated.

Prompting Tips

HoochMix MOSS works with both short natural-language prompts and detailed descriptions.

To encourage anime output, try:

anime illustration, detailed anime character, fantasy outfit, cinematic lighting, vibrant colors

To encourage semi-realistic output, try:

semi-realistic character, natural proportions, detailed skin, soft cinematic lighting

To encourage realistic output, try:

realistic photography, natural skin texture, daylight, realistic proportions, photographic lighting

Style tags are optional. The model can sometimes produce realistic-looking images without them.

A Prompt Enhancer is not required. Several successful generations were obtained directly from simple prompts without enhancement.

Important Notes

• This is an experimental community checkpoint merge.

• It is designed primarily for image quality and stylistic versatility rather than maximum generation speed.

• 25 steps / CFG 2.5 are the recommended starting settings.

• The original 10-step / CFG 1 experiments produced less reliable anatomy in some complex compositions.

• Hands, feet, complex poses and dense mechanical details can still present artifacts.

• Refiner artifacts should not automatically be attributed to the base checkpoint.

• Feedback and comparison images are welcome.

I hope you enjoy HoochMix MOSS!

Feel free to report any issues, bugs, or unexpected results you encounter. Your feedback is always welcome and will help me improve future versions.

And if you enjoy using the model, I'd love to see your creations! Feel free to share your generations in the comments.

Thank you for trying HoochMix MOSS, and happy generating! ❤️