Watch the full video first if you want to understand how this Bernini-R subject replacement workflow works in practice. The video shows how a source video and one or more reference images can be combined into a video editing pipeline, where the original subject is replaced while the motion, camera framing, background, lighting, and scene structure remain close to the original clip.
This ComfyUI workflow is designed for Bernini-R subject replacement video editing. Its main purpose is to take an existing video, read the original motion and scene layout, then use reference images to replace the main person or object in the video. Instead of generating a completely new video from scratch, this workflow focuses on controlled video editing: keeping the original action, pose, timing, camera perspective, shadows, background, and environment while transferring a new reference subject into the scene.
The workflow is built around the Bernini-R video editing model route. It uses Bernini_HIGH_fp8_e4m3fn_scaled.safetensors and Bernini_LOW_fp8_e4m3fn_scaled.safetensors as the high-noise and low-noise model branches. It also uses UMT5 XXL fp8 text encoding, Wan 2.1 VAE, BerniniConditioning, LoadVideo, GetVideoComponents, BatchImagesNode, KSamplerAdvanced, VAEDecode, CreateVideo, SaveVideo, and VHS_VideoCombine. The graph also includes UnifiedReward-Flex LoRA for both high and low model branches, helping the result maintain stronger visual quality and reward-aligned output.
The key node is BerniniConditioning. This node receives the source video, reference images, positive and negative text conditioning, VAE, width, height, video length, and reference maximum size. In this workflow, the source video provides the motion and scene base, while the reference image batch provides the new subject identity. This is the central logic behind subject replacement: the video decides how the subject moves, while the reference image decides who or what appears in the edited result.
A major advantage of this workflow is the built-in prompt creation chain. The graph uses RHLLMChatNode to analyze the uploaded video and reference image, then generate a more precise Bernini editing prompt. The LLM output is passed through a JSON cleanup chain using StringReplace nodes, then automatically connected into the positive prompt encoder. This reduces the need to manually write long editing instructions every time.
The workflow also includes task-type prompt prefix guidance. Bernini-R supports different trained task modes such as video editing, reference-based video editing, subject-to-video generation, ads insertion, style editing, motion adjustment, and content propagation. This makes the system more flexible than a simple V2V replacement graph.
Compared with ordinary video-to-video workflows, this Bernini-R workflow is more specialized for subject replacement. A normal V2V workflow may change the whole scene or lose the original motion. This workflow is designed to preserve the source video structure while replacing the target subject with a reference-guided identity. It is suitable for dance video replacement, character replacement, model try-on style video edits, advertising character insertion, AI cosplay videos, short-form video edits, product demonstration edits, Bilibili showcases, YouTube tutorials, RunningHub releases, and Civitai workflow publishing.
Main features:
Bernini-R subject replacement video editing workflow
Source video + reference image input
Replaces the main subject while preserving original motion
Bernini HIGH / LOW fp8 dual-model route
UMT5 XXL fp8 text encoder
Wan 2.1 VAE decoding
BerniniConditioning video editing control
Multiple reference images batched in
RHLLMChatNode automatic prompt generation
JSON cleanup chain for LLM prompt output
KSamplerAdvanced segmented sampling
UnifiedReward-Flex HIGH / LOW LoRA support
CreateVideo / SaveVideo / VHS output
Suitable for V2V, R2V, RV2V, and subject replacement tasks
Suggested workflow:
Prepare a clean source video first. The subject should be visible, the movement should be readable, and the camera should not shake too aggressively. Then prepare one or more reference images for the replacement subject. Use clear full-body or half-body references when possible, with stable lighting and visible clothing details. Load the video and reference images into the workflow, then let the LLM prompt chain generate the subject replacement instruction. Check the cleaned prompt before rendering. If the subject replacement is weak, make the instruction more explicit. If the background changes too much, strengthen the preservation rules for background, lighting, shadows, camera framing, and original motion. Start with a short clip first, confirm the identity and motion transfer, then move into longer video output.
⚙️ RunningHub Workflow
Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/post/2062503702258348033?inviteCode=rh-v1111
If the results meet your expectations, you can later deploy it locally for customization.
🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!
📺 Bilibili Updates (Mainland China & Asia-Pacific)
If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.
📺 Bilibili Video: https://www.bilibili.com/video/BV1yLEc6dEJc/
☕ Support Me on Ko-fi
If you find my content helpful and want to support future creations, you can buy me a coffee ☕.
Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.
👉 Ko-fi: https://ko-fi.com/aiksk
💼 Business Contact
For collaboration or inquiries, please contact aiksk95 on WeChat.
⚙️打开下方链接即可在线体验，无需安装。
👉 工作流： https://www.runninghub.ai/post/2062503702258348033?inviteCode=rh-v1111
如果觉得效果理想，你也可以在本地进行自定义部署。
🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！
📺 Bilibili 更新（中国大陆及南亚太地区）
如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。
📺 B站视频： https://www.bilibili.com/video/BV1yLEc6dEJc/
我会在 夸克网盘 持续更新模型资源：
👉 https://pan.quark.cn/s/20c6f6f8d87b
这些资源主要面向本地用户，方便进行创作与学习。