The official ComfyUI workflows for Verboa Image 1.0, the photoreal 8-step ERNIE-Image fine-tune for adults (model: https://civarchive.com/models/2962499).Each version of this page is one workflow: NVIDIA fp8, GGUF (smaller cards, needs ComfyUI-GGUF), Mac (bf16/int8 on Apple Silicon), and the two Prompt Writer workflows, where Verboa Prompt Writer writes the prompt for you (one custom node, on GitHub: https://github.com/verboa/Verboa-Image).Everything else is stock ComfyUI, since ERNIE-Image is supported natively: 8 steps, CFG 2, euler with the simple scheduler, ModelSamplingSD3 at shift 4.0, and an empty CLIPTextEncode as the negative (not ConditioningZeroOut). Besides the model file you need ministral-3-3b.safetensors (models/text_encoders) and flux2-vae.safetensors (models/vae) from Comfy-Org/ERNIE-Image on Hugging Face.All formats of the model, including GGUF from 4.8 GB, NF4 and MLX for mflux: <a target="_blank" rel="ugc" href="https://huggingface.co/verboa/Verboa-Image-1.0">https://huggingface.co/verboa/Verboa-Image-1.0</a>Same license as the model: CreativeML Open RAIL++-M plus one added restriction, no sexual or intimate images of a real, identifiable person without their consent. Adults only (18+).gingface.co/verboa/Verboa-Image-1.0Same license as the model: CreativeML Open RAIL++-M plus one added restriction, no sexual or intimate images of a real, identifiable person without their consent. Adults only (18+).

Verboa Image 1.0 · ComfyUI Workflows (fp8 / GGUF / Mac / Prompt Writer)

For RTX 40/50 and other NVIDIA cards: load verboa-image-1.0-fp8.safetensors (models/diffusion_models), ministral-3-3b.safetensors (models/text_encoders) and flux2-vae.safetensors (models/vae). Opens in App Mode with prompt, size, seed, steps and CFG as the only controls. About 20-27 s per 1 MP image on a laptop RTX 5090.

Files with SHA256: 3a6f7d733f828b0fa6098134724f61e6aee718a2ce12da47f69f51a9f948c1b3