ComfyUI-Inference-Core-Nodes
LykosAI
ComfyUI-Inference-Core-Nodes enhances COMFYUI by providing custom nodes for improved inference capabilities, streamlining workflows and boosting performance through optimized ONNX runtime options.
39
2025-11-28
Z-Image Turbo with Controlnet 2.1 and Qwen VLM for Creating Accurate Variety of Images
API
Controlnet
Floyo API
Image2Image
LoRA
Z-Image Turbo
2
280
Overview
Z-Image Turbo is a 6B diffusion transformer that generates high‑quality images in a few steps and works well with ControlNet nodes in ComfyUI for pose, depth, or edge guidance. ControlNet (via DWPose, Canny, Depth, or Union) lets you lock in composition, pose, and structure from a reference image while still using text to define style and details. Qwen VLM (Qwen-VL / Qwen2.5-VL) is a vision‑language model that can analyze images, describe them, refine prompts, or validate whether the generated image matches your textual intent.
Who can use it
This combo is useful for:
Creators who want consistent characters and poses across many images, using ControlNet for structure and Qwen VLM to keep prompts and outputs on‑brief.
Designers and marketers needing accurate branded visuals, where Qwen VLM checks logos, colors, or layout while Z-Image Turbo + ControlNet keep layout fixed.
ComfyUI power users building complex graphs that mix text‑to‑image, reference guidance, and VLM‑driven prompt refinement for higher reliability.
Anyone doing dataset creation or concept exploration who wants an automated loop: generate → analyze with Qwen VLM → adjust prompt or ControlNet → regenerate.
Use case workflow
A typical workflow is:
Feed a pose or layout image into ControlNet (DWPose, Depth, or Canny) and write an initial prompt for Z-Image Turbo, then generate a first batch of guided images.
Send one or more outputs to Qwen VLM, ask it to describe the image or compare it to your intended description, and use its detailed text as an improved prompt or prompt expansion.
Regenerate with Z-Image Turbo + ControlNet using the refined prompt, optionally repeating the loop until Qwen VLM’s analysis says the image closely matches the target concept, pose, and details.
This way, Z-Image Turbo provides speed and quality, ControlNet provides spatial accuracy, and Qwen VLM provides semantic accuracy, giving you a robust system for creating a wide variety of precise, repeatable images.
Read more
N
EXTENSIONS
Nodes & Models
Overview
Z-Image Turbo is a 6B diffusion transformer that generates high‑quality images in a few steps and works well with ControlNet nodes in ComfyUI for pose, depth, or edge guidance. ControlNet (via DWPose, Canny, Depth, or Union) lets you lock in composition, pose, and structure from a reference image while still using text to define style and details. Qwen VLM (Qwen-VL / Qwen2.5-VL) is a vision‑language model that can analyze images, describe them, refine prompts, or validate whether the generated image matches your textual intent.
Who can use it
This combo is useful for:
Creators who want consistent characters and poses across many images, using ControlNet for structure and Qwen VLM to keep prompts and outputs on‑brief.
Designers and marketers needing accurate branded visuals, where Qwen VLM checks logos, colors, or layout while Z-Image Turbo + ControlNet keep layout fixed.
ComfyUI power users building complex graphs that mix text‑to‑image, reference guidance, and VLM‑driven prompt refinement for higher reliability.
Anyone doing dataset creation or concept exploration who wants an automated loop: generate → analyze with Qwen VLM → adjust prompt or ControlNet → regenerate.
Use case workflow
A typical workflow is:
Feed a pose or layout image into ControlNet (DWPose, Depth, or Canny) and write an initial prompt for Z-Image Turbo, then generate a first batch of guided images.
Send one or more outputs to Qwen VLM, ask it to describe the image or compare it to your intended description, and use its detailed text as an improved prompt or prompt expansion.
Regenerate with Z-Image Turbo + ControlNet using the refined prompt, optionally repeating the loop until Qwen VLM’s analysis says the image closely matches the target concept, pose, and details.
This way, Z-Image Turbo provides speed and quality, ControlNet provides spatial accuracy, and Qwen VLM provides semantic accuracy, giving you a robust system for creating a wide variety of precise, repeatable images.
Read more
N
EXTENSIONS