ministral-3-3b-TE-only-ComfyUI

Text encoder extracted and pruned from Comfy-Org/ERNIE-Image, where it is distributed as ministral-3-3b.safetensors.

The base model is mistralai/Ministral-3-3B-Base-2512. This repository contains a pruned version with the vision layers and multimodal projection heads removed, retaining only the components needed for text encoding: token embeddings, transformer layers, and the final layer norm.

This model works as a drop-in text encoder for pipelines that consume the Comfy-Org/ERNIE-Image text encoder, without the overhead of the full multimodal model. It's primarily intended for use with ComfyUI workflows that load the text encoder separately using the Load CLIP node.

Prompt

This is a photograph depicting an urban street scene. Shot at eye level, it shows a covered pedestrian or commercial street.
Slightly below the center of the frame, a cyclist rides away from the camera toward the background, appearing as a dark silhouette
against backlighting with indistinct details. The ground is paved with regular square tiles, bisected by a prominent tactile paving
strip running through the scene, whose raised textures are clearly visible under the light. Light streams in diagonally from the right
side of the frame, creating a strong backlight effect with a distinct Tyndall effect—visible light beams illuminating dust or vapor in
the air and casting long shadows across the street. Several pedestrians appear on the left side and in the distance, some with their backs
to the camera and others walking sideways, all rendered as silhouettes or semi-silhouettes. The overall color palette is warm, dominated
by golden yellows and dark browns, evoking the atmosphere of dusk or early morning.

The following information was pruned from the original model:

• Vision encoder layers
• Multimodal projection heads
• PixtralProcessor references

The transformer backbone and remain untouched relative to the original ministral-3-3b.safetensors from Comfy-Org.

Files

📂 ministral-3-3b-TE-only-ComfyUI/
├── config.json              # Model configuration
├── tokenizer.json           # Tekken tokenizer vocabulary
├── tokenizer_config.json    # Tokenizer configuration
├── model.safetensors        # Pruned text encoder weights (bfloat16)
└── README.md

License

This model is released under the Apache 2.0 license, the same as the base model mistralai/Ministral-3-3B-Base-2512. The pruning process does not introduce any additional restrictions.