OpenBlender

Agent harness inside Blender — TUI floating overlay (opencode/hermes-style), not a simple chat.

1. Enable the Chat panel in addon Preferences
2. Configure provider: OpenRouter (API key) or Unsloth Studio (local, auto-discovers port & model)
3. Click the floating chat button — statusline shows model · ctx used/total (%) · compactions

Harness notes: thinking streams live as ▾ reasoning (rolling last-10 tail), then collapses to ▸ reasoning per message (global think toggle). Long tasks auto-compact at 80% context, queue while busy, and run up to 200 tool rounds with stall detection — cancel anytime. Unsloth Studio model/reload is picked up automatically.

Vision & Image Attachments

The AI chat supports vision-capable models for analysing images directly inside Blender. Vision is auto-detected — screenshot/viewport tools are only offered to the agent when the model actually accepts images.

1. Attach an image to any prompt (attachment chip or Ctrl+V in the floating chat)
2. Or ask the agent to look at Blender: "take a screenshot and tell me what's wrong with my lighting"
3. The agent can also read the current Image Editor image (capture_image_editor) or the viewport (capture_viewport)

Requires a vision-capable model. Images are sent as compressed JPEG to stay within context limits.

Batch Generation

Request multiple images at once for keyframe sequences:

• "Generate 3 continuous keyframes of Batman — first, middle, and end frames"
• "Create 5 variations of a mountain landscape"

Chat can be used to rewrite / refine your prompts.

Generate images from your viewport.

Manual Mode

1. Set mode to "Run" in sidebar
2. Adjust parameters → click "Generate"

On-Change Mode

1. Set mode to "Run (on change)" → click "Start"
2. Modify viewport — generation triggers when inputs are released

Output: Images appear in Image Editor as Result, Result_001, Result_002…

Generate videos from a starting image using LTX 2.5.

1. Display the start image in the Image Editor
2. Open the IMG to VID panel (N key → OpenBlender tab)
3. Enter a motion prompt, set Frame Count and Megapixels
4. Click "Generate Video"

The generated video auto-adds to the Video Sequencer if the scene is selected.

AI Video Prompt Generation

Ask the AI Chat to write a motion-focused video prompt from your start image — with a vision-capable model it can see the current Image Editor image and describe subject motion, camera motion, and lighting changes.

Generate videos from text, image, or video inputs using LTX 2.5.

Text to Video

1. Open the OpenBlender sidebar panel (image editor) (N key → OpenBlender tab)
2. Select the TXT to VID pannel
3. Enter a text prompt describing the scene and motion
4. Click "Generate"

IC-LoRA Depth Control

The IC-LoRA depth model gives LTX 2.5 structural guidance for consistent motion across frames. Models are downloaded automatically on first use.

Convert 2D images to 3D GLB models via Pixal3D.

1. Import an image
2. Click "Start"
3. 3D model imports into Blender as GLB, with textures

Transform images from inside the Image Editor side panel.

1. Open any image in the Image Editor
2. Press N → find the OpenBlender tab in the sidebar
3. Set your Prompt, Megapixels, and Steps
4. Click "Generate"
5. Result appears as IMG_to_IMG.001, IMG_to_IMG.002… in the Image Editor

MCP / AI Chat

IMG to IMG is also available via the MCP server and AI Chat using the trigger_remix tool. Agents can transform any displayed image remotely with a text prompt.

Image Capture

Request the current state as an image:

• "Send me the current image" → Agent calls capture_image_editor
• "Show me the viewport" → Agent calls capture_viewport

Both return base64 JPEG images that the agent can display in the chat.

Generate 360° equirectangular HDRI environment maps from text prompts using FLUX.2-Klein + Klein 9B HDRI LoRA.

1. Set viewport shading to Material Preview or Rendered so the environment is visible
2. In Properties → World, ensure a World data block exists — click "New" if the slot is empty
3. Open the OpenBlender sidebar panel (N key → OpenBlender tab)
4. Enter a Prompt describing the environment
5. Set Megapixels and Steps
6. Click "Generate HDRI"
7. The generated panorama is automatically applied to the scene World as an environment texture

Generate a humanoid dummy rig with motion from a text prompt using HY-Motion. The output is a generic animated skeleton — not a character with specific appearance or textures.

1. In Blender, open the sidebar (N) → OpenBlender tab → TXT to RIG panel
2. Enter a text prompt describing the motion (e.g. "walking forward", "jumping", "waving hand")
3. Click "Generate" — HY-Motion runs in ComfyUI and produces an animated GLB
4. The humanoid dummy rig is automatically imported into the scene

Tips

• Prompts should describe motion only — HY-Motion outputs a generic humanoid skeleton regardless of character description
• Use it as an animation base: retarget the motion onto your own character rig
• The AI Chat can trigger generation via the trigger_txt_to_rig MCP tool
• Models are downloaded from Blender Preferences

Add LoRA models to your generation workflows.

1. Open your IMG to IMG or IMG to VID workflow in ComfyUI WebUI
2. Add your LoRA node(s) to the workflow graph
3. Save the API workflow as .json
4. Your LoRA stack is now applied automatically in Blender on every generation

Any ComfyUI-compatible LoRA works. Stack multiple LoRAs in a single workflow. Find models at CivitAI.

Generate a rigged humanoid from the image shown in the Image Editor.

1. Enable the IMG to RIG (Humanoid) panel in addon Preferences
2. Display a humanoid image in the Image Editor
3. Open the sidebar → OpenBlender → IMG to RIG (Humanoid)
4. Click Generate Rig from Image — rigged GLB imports into scene

Generate rigged animation from video.

1. Enable the VID to RIG panel in addon Preferences
2. Open the sidebar → OpenBlender → VID to RIG
3. Load a video file showing human motion
4. Click Generate — animated rig imports into scene