ltx-2.3-10eros-v1.2-mlx-q8

LTX-2.3 10Eros v1.2 — MLX int8 (q8)

Int8 (8-bit) MLX quantization of TenStrip/LTX2.3-10Eros v1.2.

Converted with mlx-forge and packaged for ltx-2-mlx — a pure-MLX port of LTX-2 for Apple Silicon.

[!WARNING] Not for all audiences. 10Eros is intended for adult use. By downloading and running this model you confirm you are of legal age in your jurisdiction and accept responsibility for the content you generate. Do not use it to produce illegal material or to depict real, identifiable people without consent.

---

What's in here

This repo ships the full split layout that ltx-2-mlx expects — both transformer variants plus all shared components, so two-stage, distilled, I2V and audio pipelines all work out of the box.

Quantization: int8, group size 64, applied only to nn.Linear inside transformer_blocks. AdaLN, projections, connectors, VAE and vocoder remain bf16 (MLX cannot quantize Conv layers).

Text encoder: Gemma 3 12B is not bundled — ltx-2-mlx loads it separately via mlx-lm.

---

Usage

Requires ltx-2-mlx on Apple Silicon.

# Text-to-video (distilled, fastest)
ltx-2-mlx generate \
  --model /path/to/ltx-2.3-10eros-v1.2-mlx-q8 \
  --prompt "your prompt" \
  --distilled \
  -H 480 -W 704 -f 97 -o out.mp4

# Two-stage (dev + CFG + upscale, recommended quality)
ltx-2-mlx generate \
  --model /path/to/ltx-2.3-10eros-v1.2-mlx-q8 \
  --prompt "your prompt" \
  --two-stage -o out.mp4

# Image-to-video — add --image to any mode
ltx-2-mlx generate \
  --model /path/to/ltx-2.3-10eros-v1.2-mlx-q8 \
  --prompt "animate this" \
  --two-stage --image photo.jpg -o out.mp4

On 32 GB Macs add --low-ram for block-streamed inference. q8 fits 16 GB with --low-ram.

---

Conversion provenance

1. The cond-safe distilled LoRA was merged into the 10Eros v1.2 bf16 base to produce a distilled bf16 checkpoint:

   uv run ~/mlx-forge/scripts/merge_lora.py
   # → /Volumes/Storage/10eros-v1.2-distilled-bf16.safetensors
   

2. Distilled variant + all shared components + upscalers, quantized to int8:

   uv run mlx-forge convert ltx-2.3 --variant distilled-1.1 \
     --checkpoint /Volumes/Storage/10eros-v1.2-distilled-bf16.safetensors \
     --quantize --bits 8 \
     --spatial-upscaler x2 x1.5 --temporal-upscaler x2 \
     --output models/ltx-2.3-10eros-v1.2-mlx-q8
   

3. Dev variant (10Eros v1.2 base) added into the same dir, sharing the components above:

   uv run mlx-forge convert ltx-2.3 --variant dev \
     --checkpoint ~/mlx-forge/models/10Eros_v1.2_bf16.safetensors \
     --quantize --bits 8 --skip-shared \
     --output models/ltx-2.3-10eros-v1.2-mlx-q8
   

---

Credits & license

• Finetune: TenStrip/LTX2.3-10Eros
• Base model: Lightricks/LTX-2.3
• Conversion: mlx-forge · Runtime: ltx-2-mlx

Distributed under the LTX-2 license. The 10Eros finetune's own terms also apply — review the upstream model card before redistribution.