RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2089672439675580417?inviteCode=rh-v1111
Anima 2.9B Text-to-Image + Face and Hand Repair
Purpose: generate an Anima 2.9B image, then automatically detect and re-render faces and hands with detailers. This is the workflow for the usual anime failure modes: hands with the wrong number of fingers, faces that lose the design at small scale.
Graph (as saved)
29 nodes / 45 links.
- UNETLoader
Anima-2.9B-preview-v1.safetensors; CLIPLoaderqwen_3_06b_base.safetensors(typestable_diffusion); VAELoaderKJqwen_image_vae.safetensors. - Base: EmptyLatentImage 512x512, FluxResolutionNode 1.0 MP; KSampler 28 steps / CFG 5 /
res_multistep/linear_quadratic/ denoise 1, thenLatentUpscaleBy1.5x and a second KSampler at denoise 0.35. - Detailers: two
FaceDetailernodes (one at 1024, one at 832 blocks) driven by threeUltralyticsDetectorProvidermodels —bbox/face_yolov8n.pt,bbox/hand_yolov8s.ptandsegm/PitHandDetailer-v2-Test-v9c.pt— plusSAMLoadersam_vit_b_01ec64.pth(AUTO). Detailer sampling: 28 steps / CFG 5 / res_multistep / linear_quadratic at denoise 0.30 / 0.35. - Two SaveImage outputs:
Anima_2.9B/base_previewandAnima_2.9B/face_hand_refined.
How to use
The detectors are the part most likely to need attention: if a hand is not detected, no repair happens, and the graph will not warn you. Check both saved outputs. The detailer denoise values (0.30 / 0.35) are a balance between fixing anatomy and changing the design.
Limits and not-verified items
No inference was run for this listing: no speed, VRAM, detection-rate or repair-quality claim is made; values are quoted from the saved nodes. Detector and SAM weights are not bundled and have their own licenses. The cover image is a promotional still, not a frame produced by this graph.
Tutorial: YouTube · Bilibili. Workflows supplied through AI-KSK; verify workflow, node, model, detector, SAM, image and cover rights separately. No blanket commercial or redistribution license is asserted.
Description
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2089672439675580417?inviteCode=rh-v1111
Anima 2.9B Text-to-Image + Face and Hand RepairPurpose: generate an Anima 2.9B image, then automatically detect and re-render faces and hands with detailers. This is the workflow for the usual anime failure modes: hands with the wrong number of fingers, faces that lose the design at small scale.
Graph (as saved)29 nodes / 45 links.
- UNETLoader
Anima-2.9B-preview-v1.safetensors; CLIPLoaderqwen_3_06b_base.safetensors(typestable_diffusion); VAELoaderKJqwen_image_vae.safetensors. - Base: EmptyLatentImage 512x512, FluxResolutionNode 1.0 MP; KSampler 28 steps / CFG 5 /
res_multistep/linear_quadratic/ denoise 1, thenLatentUpscaleBy1.5x and a second KSampler at denoise 0.35. - Detailers: two
FaceDetailernodes (one at 1024, one at 832 blocks) driven by threeUltralyticsDetectorProvidermodels —bbox/face_yolov8n.pt,bbox/hand_yolov8s.ptandsegm/PitHandDetailer-v2-Test-v9c.pt— plusSAMLoadersam_vit_b_01ec64.pth(AUTO). Detailer sampling: 28 steps / CFG 5 / res_multistep / linear_quadratic at denoise 0.30 / 0.35. - Two SaveImage outputs:
Anima_2.9B/base_previewandAnima_2.9B/face_hand_refined.
The detectors are the part most likely to need attention: if a hand is not detected, no repair happens, and the graph will not warn you. Check both saved outputs. The detailer denoise values (0.30 / 0.35) are a balance between fixing anatomy and changing the design.
Limits and not-verified itemsNo inference was run for this listing: no speed, VRAM, detection-rate or repair-quality claim is made; values are quoted from the saved nodes. Detector and SAM weights are not bundled and have their own licenses. The cover image is a promotional still, not a frame produced by this graph.
Tutorial: YouTube · Bilibili. Workflows supplied through AI-KSK; verify workflow, node, model, detector, SAM, image and cover rights separately. No blanket commercial or redistribution license is asserted.