<div data-youtube-video="https://youtu.be/eQep5Vq03RA"><iframe src="https://www.youtube.com/embed/eQep5Vq03RA" width="853" height="480" allowfullscreen></iframe></div> 
 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2102678363970605058?inviteCode=rh-v1111Bilibili tutorial video: https://www.bilibili.com/video/BV14cht6NE8D/# Qwen Image 2.1 — Character Pose Direction## Purpose and saved setupGuide character generation with a Pose Studio image at encoder image1 and a separate appearance reference at image2. This single14-node workflow is not an explicit hard-joint ControlNet route or a verified identity-preservation system.Studio lighting_prompt overrides the encoder's old pose-replacement widget. Three saved prompt strings say Draw character from image2. Inspect refreshed editor outputs rather than assuming the old sentence or a lighting port name establishes the actual instruction.## PreparationResolve compatible Pose Studio, Qwen encoder/cache and standard sampling nodes. Load an authorized character image, inspect and refresh the editor's poses/captures, then check the emitted prompt and effective dimensions.Three saved captures are1520×1024 by header inspection, with LIST export. Linked editor dimensions override the latent's old1024 defaults, but the empty latent retains batch1 and the batch-count output is unconnected. Verify actual list/batch behavior and save count; do not promise three completed outputs.The model passes through Pose Studio V1.1 LoRA at strength1 and a gpu/int8 cache configuration. Sampling stores25 steps, CFG1, Euler/simple, denoise1 and a fixed seed. VAE is linked to encoder and decoder, but sampling starts from an empty latent, not the encoder latent output. No cache speed or memory benefit was measured.## Limitations and outputThe comparer shows generated image versus Studio image, not the loaded character reference. Inspect identity separately. One SaveImage node uses prefix ComfyUI; actual file count requires testing. Assess pose, hands, silhouette, costume, lighting and unintended edits after a small run. Source reports of16 groups with2 unsuccessful cases are not independently accepted benchmark results.## Asset references- `qwen_image_2.1_int8_convrot.safetensors`- `qwen3vl_8b_int8_convrot.safetensors`- `qwen_image_2.1_vae_bf16.safetensors`- `VNCCS_QI2_PoseStudioV1.1.safetensors`## Files and version note- [Workflow JSON]()- [Selected cover]()The historical comparer gallery is cleared in new copies; complete Pose Studio state and embedded captures are retained because they may affect execution. Review embedded media content and rights before publishing. Source originals are unchanged. Selected covers are not verified output samples. No model run, visual acceptance, timing, memory or license verification is included. This local description is not a remote release or model upgrade.## Tutorial and rights[Source-provided tutorial](https://www.youtube.com/watch?v=eQep5Vq03RA), not live-verified. The workflow is supplied through AI-KSK.Model weights and the selected character-image file are not bundled; the workflow does retain editor captures. Check workflow, node, model, LoRA, editor-media, character-image and cover permissions separately. No blanket commercial license or pose-fidelity guarantee is asserted.

Qwen Image 2.1 Character Pose Director (VNCCS Pose Studio)

<div data-youtube-video="https://youtu.be/eQep5Vq03RA"></div> 
 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2102678363970605058?inviteCode=rh-v1111Bilibili tutorial video: https://www.bilibili.com/video/BV14cht6NE8D/# Qwen Image 2.1 — Character Pose Direction## Purpose and saved setupGuide character generation with a Pose Studio image at encoder image1 and a separate appearance reference at image2. This single14-node workflow is not an explicit hard-joint ControlNet route or a verified identity-preservation system.Studio lighting_prompt overrides the encoder's old pose-replacement widget. Three saved prompt strings say Draw character from image2. Inspect refreshed editor outputs rather than assuming the old sentence or a lighting port name establishes the actual instruction.## PreparationResolve compatible Pose Studio, Qwen encoder/cache and standard sampling nodes. Load an authorized character image, inspect and refresh the editor's poses/captures, then check the emitted prompt and effective dimensions.Three saved captures are1520×1024 by header inspection, with LIST export. Linked editor dimensions override the latent's old1024 defaults, but the empty latent retains batch1 and the batch-count output is unconnected. Verify actual list/batch behavior and save count; do not promise three completed outputs.The model passes through Pose Studio V1.1 LoRA at strength1 and a gpu/int8 cache configuration. Sampling stores25 steps, CFG1, Euler/simple, denoise1 and a fixed seed. VAE is linked to encoder and decoder, but sampling starts from an empty latent, not the encoder latent output. No cache speed or memory benefit was measured.## Limitations and outputThe comparer shows generated image versus Studio image, not the loaded character reference. Inspect identity separately. One SaveImage node uses prefix ComfyUI; actual file count requires testing. Assess pose, hands, silhouette, costume, lighting and unintended edits after a small run. Source reports of16 groups with2 unsuccessful cases are not independently accepted benchmark results.## Asset references- `qwen_image_2.1_int8_convrot.safetensors`- `qwen3vl_8b_int8_convrot.safetensors`- `qwen_image_2.1_vae_bf16.safetensors`- `VNCCS_QI2_PoseStudioV1.1.safetensors`## Files and version note- [Workflow JSON]()- [Selected cover]()The historical comparer gallery is cleared in new copies; complete Pose Studio state and embedded captures are retained because they may affect execution. Review embedded media content and rights before publishing. Source originals are unchanged. Selected covers are not verified output samples. No model run, visual acceptance, timing, memory or license verification is included. This local description is not a remote release or model upgrade.## Tutorial and rights[Source-provided tutorial](https://www.youtube.com/watch?v=eQep5Vq03RA), not live-verified. The workflow is supplied through AI-KSK.Model weights and the selected character-image file are not bundled; the workflow does retain editor captures. Check workflow, node, model, LoRA, editor-media, character-image and cover permissions separately. No blanket commercial license or pose-fidelity guarantee is asserted.

Files with SHA256: ae0fed2e8bdb6f26e39b32344b3e47c07883b94caad57ddba0faebfb40f18383