Text to Image

Text-to-image pipelines built on Fibo 2.0, Flux 3 Klein, Qwen Image 2.1 Turbo, Qwen Image 2.1, Ideogram 4, and Krea 2 GGUF quants.

Models and formats

The following models support Text-to-Image modalities:

Model	Strength	Tradeoff
Fibo 2.0 Dev	Text-to-image and image editing with FLUX.2 latents	Requires ComfyUI core at the specified minimum commit and Qwen3-VL 4B
Flux 3 Klein	Compact 5.6B model with layout control and text rendering	Requires Qwen3-VL 4B text encoder and ComfyUI 0.41.0+
Ideogram 4	Excellent prompt following, great text rendering	Strict JSON prompt structure is cumbersome to follow.
Krea 2	Highly creative, decent prompt following	Text rendering not always perfect
Qwen Image 2.1 Turbo	Fast text-to-image generation and image editing, designed for 8 denoising steps	Requires the Qwen3-VL 8B text encoder
Qwen Image 2.1	Text-to-image with optional image references	Requires the Qwen3-VL 8B text encoder

Loading instructions

1. Download the required GGUF files from Hugging Face and put it in the models/diffusion_models ComfyUI directory. You only need one quant format, not all.
2. Replace the default diffusion model node with the GGUF loader.
3. Point the loader at the model's .gguf file.
4. Load the pruned text encoder GGUF alongside it.
5. Set the sampler steps and CFG to the model's recommended values. (Using a GGUF instead of a safetensors/other file format does not affect what sampling/scheduling/etc settings you need to set.)