OmniGen2: Text to Image
Generate high-quality images from text prompts using OmniGen2's unified 7B multimodal model with dual-path architecture.
About
This workflow demonstrates a minimal, fast path from text prompt to finished image using OmniGen2’s unified 7B multimodal model. A single OmniGen2 node (custom node id 4b17d220-4312-4981-9eae-9a76bf3b6ec9) ingests your prompt, encodes it with the Qwen2.5-VL text encoder, and produces a high-quality image without the complexity of diffusion samplers or long node chains. The output is then passed to SaveImage for consistent file naming and export, while a MarkdownNote provides in-canvas guidance and links to the official docs.
What you provide
- Reading the workflow
Runs in ComfyUI. This workflow can't run on this page. Open it in Comfy Cloud with everything loaded, or download it for your own ComfyUI.
Try in CloudQuestions & answers
Place qwen_2.5_vl_fp16.safetensors in ComfyUI/models/text_encoders. If you’re using the repackaged OmniGen2 weights from Hugging Face, follow their folder structure and naming exactly.
Set width and height in the OmniGen2 node. Common pairs are 1024x1024 (square), 1344x768 (16:9), or 832x1216 (2:3). Higher resolutions require more VRAM, so increase gradually.
Use a fixed seed value in the OmniGen2 node. To explore variations, either change the seed slightly or enable batch/count to generate multiple outputs in one pass.
If your OmniGen2 node exposes a negative_prompt input, use it to steer away from undesired elements. If not, phrase constraints directly in your prompt (e.g., “no text, minimal background, without watermark”).