Abiray/10Eros-Max-GGUF

10Eros-Max (GGUF Quants)

This repository contains highly optimized GGUF quantizations of TenStrip/10Eros-Max (Test4 Pruned).

These files allow users to run this massive 40GB+ experimental multi-model merge on consumer hardware with as little as 12GB of VRAM, significantly reducing memory overhead while preserving the model's unique visual fidelity, temporal coherence, and prompt adherence.

---

📦 Available Files & VRAM Guide

File Name	Size	Target Hardware	Recommended VRAM	Description
10Eros-Max-Test4-Pruned-Q8_0.gguf	21.6 GB	RTX 3090 / 4090 / Mac M-Series	24GB+	The purist's choice. Visually lossless to the original BF16 model. Perfect for high-end workstations with ample VRAM.
10Eros-Max-Test4-Pruned-Q6_K.gguf	16.7 GB	RTX 3090 / 4090	24GB	The optimal "sweet spot." Near-imperceptible quality loss while leaving enough VRAM overhead to run high-res text encoders natively.
10Eros-Max-Test4-Pruned-Q5_K_M.gguf	14.1 GB	RTX 4080 / 4070 Ti Super	16GB – 20GB	The mainstream high-end choice. Excellent quality-to-VRAM ratio, maximizing visual fidelity for 16GB cards.
10Eros-Max-Test4-Pruned-Q4_K_M.gguf	11.6 GB	RTX 4070 / 4080	16GB	The golden standard. Significant compression that preserves the delicate cross-model stylings perfectly.
10Eros-Max-Test4-Pruned-Q3_K_M.gguf	8.9 GB	RTX 3060 / 4060	10GB – 12GB	The budget tier. Makes the model accessible to entry-level GPUs while leaving just enough room for the Qwen3-VL text encoder.

---

🚀 How to Use in ComfyUI

To run these GGUF files in ComfyUI, you must use a GGUF-compatible loader.

1. Install ComfyUI-GGUF: Install the City96/ComfyUI-GGUF custom node pack via the ComfyUI Manager.
2. Download the File: Download your preferred .gguf file from the list above and place it in your ComfyUI/models/unet/ folder.
3. Load the Model: In your workflow, replace the standard Load Checkpoint or Load Diffusion Model node with the Unet Loader (GGUF) node. Select the 10Eros-Max GGUF file.
4. Text Encoders: You will still need to load the separate text encoders natively (the H3 truncated version of Qwen3-VL-32B) and the MiniMax VAE to decode the final video/audio.

---

📖 About the Original 10Eros-Max Model

(The following is from the original creator, TenStrip)

So far, this release is an experiment on MiniMax H3 base model that combines learned patterns from two other video diffusion models — LTX 2.3 and Wan 2.2, and in test3 and later, one image diffusion model — Krea 2 — into H3's transformer architecture.

A unified transformer with modality-specific input projectors (video, audio, condition) feeding into a 52-block transformer stack (2-block token_refiner for text conditioning plus 50 main blocks), with modality-specific output heads. The transformer stack uses identical block architecture throughout, which is what makes the attention and MLP layers architecturally-appropriate targets for character transfer between models — modality routing lives at the pipeline edges, not within the transformer.

The LTX pass was applied first to H3's front blocks, followed by a Wan pass targeting H3's middle-to-back blocks. Each pass modified attention weights directly; the second pass also modified the feed-forward layers in a broader block range. Together, these two video-donor passes demonstrated that completely different video model architectures could each contribute their own recognizable character into H3.

A third pass then brought in Krea 2 — an image diffusion model rather than a video model. This was a more speculative direction, since image models learn spatial relationships within a single frame while video models learn spatiotemporal relationships across many frames. Krea's contribution was scoped narrowly to the parts of its attention structure that carry that spatial learning: the query projection (which encodes what visual features to attend to) and, in a smaller number of positions, the paired key and value projections (which encode how those features relate). The feed-forward layers were partially transferred as well, but only the input side of Krea's gated MLP — the output projection was skipped because Krea's combining logic was trained for an image-model output pathway that doesn't match H3's video processing. This narrower scope let Krea contribute finer visual character and detail work without disrupting H3's temporal coherence.

The Orthogonal Math

The modifications use orthogonal projection math — new weight patterns are added into H3 primarily in directions the base model wasn't already using. This lets the transferred character supplement H3 rather than override it. For the cross-domain Krea pass, this was especially important: Krea's patterns and H3's patterns turned out to be nearly perpendicular in weight space, giving orthogonal projection maximum room to work while preserving H3's own directions almost entirely.

Combined with careful control over which blocks are affected — and, for Krea specifically, careful control over which attention head positions receive the injection to avoid disturbing the head slots most involved in audio processing — this preserves H3's core capability to produce coherent video with audio while shifting its aesthetic and motion character toward the donor models. The model's actual block weights, safety features, and guardrails remain unmodified in the architectural sense — every modification is a controlled additive perturbation of existing weights, not a replacement of the model's own structure.

---

⚖️ License & Credits

• Base Architecture: MiniMax-H3
• Model Merge/Creation: TenStrip
• GGUF Quantization Scripting & Conversion: Abiray
• License: The standard MiniMax H3 community license applies. Because this release carries transferred character from LTX 2.3, Wan 2.2, and Krea 2, the community licenses for those source models apply as well to the portions of character that came from each.