Load one photo and MiniMax H3 films a camera orbit all the way around the subject. The same photo is the first and the last frame, so the clip ends where it starts and loops. The clip has no sound.

This workflow runs the 360 Orbit LoRA by pablodawson. All credit for the orbit goes to them.

Why use this workflow

You can load the LoRA into a plain first + last frame graph. This workflow adds four things:

• It sets the LoRA up for you. Your photo goes in as the first and the last frame, with the LoRA card's own prompt. There's no prompt to write.

• It fixes the "spinning on the spot" look. On a plain wall or backdrop the turn looks like the subject is spinning, because the camera has nothing to go past. Type what is around them and the clip shows that place going past. That worked on all three plain photos I tested.

• No bars. The photo loads through a crop tool and the clip takes its shape. One tap gives the square the LoRA was trained on instead, with black bars or as a crop.

• It is fast. A 4-step turbo LoRA is switched on, about 4x faster than the LoRA card's 28 steps.

How to use it

1. Load your photo.

2. Pick how many seconds. It opens on 4, which gives a cleaner background than 3.

3. Optional: type what is around them.

4. Run.

Plain wall or backdrop? Say what is around them

Type the place into What is around them. Two short sentences are enough, like "She is in a photo studio in front of a gray backdrop. Around her are light stands, a stool and a desk."

The place is made up from your words, so it won't be your real room. Leave the box empty when the photo already has a room or a street in it.

The shape of the clip

The clip takes the shape of your photo, with no bars. The LoRA was trained on square clips, so that is outside what it was trained on, but the camera went all the way round on the four photos I tried.

• Want the square? Tap 1:1 under your photo. It gets black bars to make it square.

• A square crop instead: with 1:1 lit, set Fit to crop and drag the box to what you want in the clip.

What you need

• AusBoss nodes 2.5.0 or newer. In ComfyUI-Manager, search "AusBoss". Already have them? Update first.

• The MiniMax H3 first/last-frame model, the 360 Orbit LoRA and the turbo LoRA:

  • minimax_h3_fl2va_pruned_int8_convrot.safetensors → models/diffusion_models · 21.0 GB

  • qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors → models/text_encoders · 15.7 GB

  • minimax_h3_video_vae_fp16.safetensors → models/vae · 5.21 GB

  • minimax_h3_flf2v_lora_v1.safetensors → models/loras · 155 MB

  • minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_bf16.safetensors → models/loras · 1.96 GB (the 4-step turbo LoRA)

• Comfy-Org recommends the int8_convrot model when your PyTorch is built for CUDA 13 (cu130), and their fp8_scaled build otherwise. Both are on the same Comfy-Org/MiniMax-H3 page.

Tips

• Turbo is on, for speed. For better quality, turn it off. Switch the turbo row off, then double-click Film the orbit and set steps to 28 and shift_video to 12. That's the LoRA card's own setting, about 4x slower. The cover clip was made that way. On a plain backdrop keep turbo on: at 28 steps the backdrop stayed flat in my tests.

• If the camera goes to the back and returns the same way, press New seed on the Seed node. Switching turbo off also fixed it on the photo where I saw it.

• In a busy room the camera passes behind things. With shelves or walls close by, it goes behind them on its way around, like the library clip on the cover.

• The LoRA was trained on 28 short clips of people, square. Other subjects and shapes are untested by its author.

• Anything the photo doesn't show gets made up, like the back of a shirt.

• Every video and picture on this page has the workflow inside. Drop one on ComfyUI to load it.

• Small movements like a blink can slip in. Press New seed and run it again.

• About 35 seconds a 4-second clip on an RTX 5090 with turbo, about two and a half minutes without.

Credits

The 360 Orbit LoRA is by pablodawson, under the MiniMax H3 Community License (its card lists the license as "other"). MiniMax H3 is by MiniMax, under the same license, and the files, including the 4-step turbo LoRA, are Comfy-Org's repackage. The license only covers its "Applicable Territory", which leaves out the European Union, the United Kingdom, South Korea and the United States, and the same limit applies to what you make with it. Commercial products above US$20M a year in revenue need MiniMax's written OK and must show "MiniMax H3".

v1.1 (2026-10-04): a box to say what is around the subject, for photos with a plain backdrop. The photo loads through a crop tool and the clip takes its shape, so there are no bars; one tap gives a square with black bars. v1.0 (2026-09-30): first release.

Questions or bugs: GitHub issues or the comments here. More from me on X @Zanzibased and GitHub.