Combined workflow (ComfyUI txt2img, Wildcards, LLMs/Ollama, Pony, SDXL, Illustrious, Flux, Qwen, Z-Image, Anima, Flux 2 Klein, Krea 2 Turbo) — v8.3

Use Combined Workflow v8.3 in ComfyUI to turn text prompts into images with SDXL, Pony, Illustrious, Flux1D, Qwen, ZImageTurbo/Base, Anima, Flux2Klein, and Krea2Turbo; optional LLMs/Ollama can extend prompts and Wildcards can be processed, producing staged 4MP and final upscaled

Stage 1 generates a regular image with the Detail Daemon sampler and sends it to a selector that can be bypassed for batch generation.

Stage 2 upscales the image to 4MP; Stage 3 uses Ultimate SD Upscale with the original model and LoRA characteristics, with optional Flux.1D resampling, Cleanup, and face, hand, and eye detailers.

Stage 4 sends the result to SeedVR2 or HighresFix for the final 16MP image, then performs color matching against the initial upscale.

Input: a text-to-image prompt, with optional prompt extension through LLMs/Ollama and Wildcards processing.

Outputs include the regular image, a 4MP intermediate upscale, a final upscaled 16MP image, and CivitAI-compatible metadata for each generation stage.

The included READ ME FIRST section covers usage details and setup requirements, including diffusion-model checkpoints and Ultralytics; published details report compatibility with ComfyUI 0.26.0.

The listed custom nodes include cg-image-filter, comfy-core, comfy-image-saver, comfy-mtb, comfyliterals, comfyui_controlnet_aux, comfyui_essentials, comfyui_llm_party, comfyui_ultimatesdupscale, comfyui-crystools, comfyui-custom-scripts, comfyui-detail-daemon, and comfyui-easy-u

They also include comfyui-fbcnn, comfyui-image-saver, comfyui-impact-pack, comfyui-inspire-pack, comfyui-kjnodes, comfyui-lora-manager, comfyui-ollama, comfyui-qwenvl, comfyui-resolution-master, comfyui-rmbg, rgthree-comfy, and seedvr2_videoupscaler.

Listed model files include 4x-UltraMix_Restore.pth, clip_l.safetensors, diffusion_pytorch_model.safetensors, jibMixFlux_v12SRPO-nvfp4.safetensors, Krea2Turbo/Krea2_Turbo_fp8mixed.safetensors, OmniSR_X2_DIV2K.safetensors, qwen_3_4b.safetensors, Qwen/jibMixQwen_v60.safetensors, Q?

For batch generation, bypass the Stage 1 selector.

If Ollama is available, set keep alive to 0 to unload the model immediately after the request; use an instruction-tuned conversion model.

Suggestion · non vérifié

A batch of two images on a 4090 is reported to take 7 minutes; use that figure for planning only, not as a guaranteed runtime.

Suggestion · non vérifié

Before use, verify that Qwen/jibMixQwen_v60.safetensors, jibMixFlux_v12SRPO-nvfp4.safetensors, qwen_3_4b.safetensors, OmniSR_X2_DIV2K.safetensors, 4x-UltraMix_Restore.pth, diffusion_pytorch_model.safetensors, Krea2Turbo/Krea2_Turbo_fp8mixed.safetensors, and Qwen/qwen_image_vae.s…

Suggestion · non vérifié