Bernini-R in ComfyUI: 4-Step LoRA, S2V Audio, VoxCPM Voice Cloning and 14 Workflows in One Mirror

Bernini-R is an open-source video model that the community has pushed hard — especially on speed. This episode packages 14 workflows, a 4-step acceleration LoRA, S2V audio and VoxCPM voice cloning into a one-click mirror. The notes below come from the episode subtitles and the published mirror page; no workflow JSON was shipped or parsed, so no node-level parameters are claimed beyond what the episode states.

Speed: the 4-step LoRA

The community produced a 4-step acceleration LoRA: two high-noise steps plus two low-noise steps. With four steps total you can generate video and apply the S2V update. The author says speed was previously the main reason exploration of this model dropped to zero; the acceleration LoRA revives it.

What the model can do

Bernini-R is a multi-function model, not just T2V/I2V:

• Base tasks: T2V, I2V.
• R2V reference-to-video, which the author especially likes and calls surprisingly strong for an open model.
• Video editing.
• With S2V, it is no longer a silent model.
• VoxCPM voice cloning is included in the mirror.
• A separate "N" version LoRA is also mentioned.

The author notes the model is a combination of LoRAs open-sourced by ByteDance, and the community nicknames it "small Jimeng".

The 14 workflows

The episode walks through all 14, and notes they are not especially complex. Two examples:

1. Video-to-video / video editing — upload a source video, set render duration and an edit instruction. The key logic is telling the model what to change and what must not change: in the background-replacement example, keep the three people's identity, clothing, formation, dance motion, camera position and original rhythm.
2. Reference editing — same as above plus one reference image; typical use is replacing a person in the source video with the character in the reference image while inheriting the original motion.

Hardware and deployment

The author recommends 50-series or above. The mirror is on 优云智算:

1. Open the mirror page: https://www.compshare.cn/images/auhhcBDqEG0N?referral_code=A7fTQUYY9lCDFJwA6MMvoq
2. Deploy an instance and wait for the image to come up.
3. Open ComfyUI, pick one of the 14 workflows and run.

Links

• Mirror (with referral code): https://www.compshare.cn/images/auhhcBDqEG0N?referral_code=A7fTQUYY9lCDFJwA6MMvoq
• Video tutorial (YouTube): https://www.youtube.com/watch?v=q_nVS3SBGwE
• Chinese original (Bilibili): https://www.bilibili.com/video/BV1nG346uEWL

Limits

• The 4-step LoRA is a community acceleration; quality is a trade-off and the episode treats it as a speed-focused route.
• Workflow count and contents are as bundled in this mirror version.
• The mirror is a deployment convenience; third-party LoRAs keep their own licenses.