Turn Any Still Photo Into a 360 Orbit Infinite Loop Video
Creating a camera orbit requires an array of synchronized DSLR cameras or a full 3D scan. With the new 360 Orbit LoRA for MiniMax H3 in ComfyUI, a single image can turn into an 360-degree camera rotation in a single generation. In this tutorial, you will learn how to: Set up the MiniMax H3…
Annons
Annons
Creating a camera orbit requires an array of synchronized DSLR cameras or a full 3D scan. With the new 360 Orbit LoRA for MiniMax H3 in ComfyUI, a single image can turn into an 360-degree camera rotation in a single generation. In this tutorial, you will learn how to: Set up the MiniMax H3 first-and-last-frame workflow in ComfyUI. Load the 360 Orbit LoRA to force circular camera rotation. Use the first-and-last-frame trick to create an infinite loop video Write a frozen-time prompt that locks every object in place while the camera moves. Software needed We will use ComfyUI, a free AI image and video generator that runs on Windows, Mac, or Google Colab. Think Diffusion provides an online ComfyUI service if you do not have a capable local GPU. (SDA readers get 20% extra credit.) What is the MiniMax H3 360 Orbit LoRA? Getting a video model to rotate 360 degrees around a subject has always been tricky. If you feed an image into the standard MiniMax H3 Image to Video, the camera might pan, but it drifts away and never returns to the starting view. What if you use the first-and-last-frame (fl2va) feature? In theory, pinning the first frame and the last frame to the exact same image should make the video loop. In practice, the base model gets lazy. Because the start and end pictures are the same, the model takes the path of least resistance: You can a static video with zero motion. That is where Pablo Dawson’s 360 Orbit LoRA steps in. Key features: Trained on 3D Gaussian Splats Enforced orbital motion Endless looping MiniMax H3 360 Orbit Step-by-Step Step 0: Update ComfyUI Make sure that your ComfyUI installation is up to date before loading the workflow. Click the Manager button on the top toolbar to open ComfyUI Manager. Click Update ComfyUI. Restart ComfyUI and refresh your browser. Step 1: Download the models Download these model files: Download minimax_h3_fl2va_pruned_int8_convrot.safetensors (19.5 GB) and put it in ComfyUI > models > diffusion_models. Download qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors (14.6 GB) and put it in ComfyUI > models > text_encoders. Download minimax_h3_video_vae_int8_convrot.safetensors (2.62 GB) and put it in ComfyUI > models > vae. Download minimax_h3_audio_vae_fp32.safetensors (577 MB) and put it in ComfyUI > models > vae. Download minimax_h3_flf2v_lora_v1.safetensors (148 MB) and put it in ComfyUI > models > loras. Make sure that you download the fl2va checkpoint, not ref2va. The fl2va model has first-and-last-frame conditioning slots that pin both ends of the generation. If minimax_h3_video_vae_fp16.safetensors is installed on your machine instead of the int8_convrot video VAE, you can select that file in the Load VAE node. Step 2: Load the workflow Download the workflow JSON file below and drop it directly onto your ComfyUI workspace: Download The pipeline is organized inside the Image to Video (MiniMax H3) subgraph. If you double-click the subgraph node, you can view the internal nodes such as model loader, LoRA patch switch, and dual VAE decoders. Step 3: Install missing nodes If any node displays a red outline after loading the workflow, install the missing dependencies: Click Manager > Install Missing Custom Nodes. Install any nodes flagged by Manager. Restart ComfyUI. Step 4: Upload your keyframe image Upload the photograph that you want to orbit into the Load Image node on the left. You can download and use this sample image generated with Flux 2 to replicate the demo: First and last frame image. Here is the key wiring trick: the single IMAGE output from Load Image connects to both the first_frame input and the last_frame input on the Image to Video (MiniMax H3) subgraph node. This sends the exact same image to both ends of the video timeline. Step 5: Configure the settings Set up generation parameters on the main nodes: Resolution: Set the ResolutionSelector node to 1:1 (Square) with 0.59 megapixels. That produces a clean 768x768 canvas, which matches the native resolution of the model. Duration: Set duration to 3.0 seconds. In MiniMax H3, the internal frame math snaps this duration to 73 frames ($17 \times 4 + 5 = 73$), which fits the model’s 24 fps latent grid. Orbit LoRA: Ensure that the LoRA switch is enabled, minimax_h3_flf2v_lora_v1.safetensors is selected in the LoRA loader, and lora_strength is set to 1.0. Sampling Steps: Set steps to 28. The res_multistep sampler with the simple scheduler produces sharp detail at 28 steps. Guidance (CFG): Leave guidance at none. MiniMax H3 is a guidance-distilled model, so it does not use a CFG slider or negative prompt. Step 6: Enter the frozen-time prompt The LoRA was trained with a specific prompt. Copy and paste this prompt into the prompt box: One frozen instant. Only the camera moves. In a continuous 360 orbit. Preserve every person and object in exactly the same world position, orientation, shape and pose throughout the shot. Airborne objects remain suspended at the captured height and angle: no wobbling, shaking, spinning, drifting, falling or continued action. Keep faces, hands, clothing, liquids and the background motionless while retaining their natural appearance. Camera parallax is the only source of apparent movement. No cuts, zoom, morphing or added objects. Keep this prompt text intact. The LoRA recognizes these exact tokens to lock the world geometry. Step 7: Generate the video Click Queue Prompt (or press Ctrl+Enter) to run the workflow. On an RTX 4090 GPU, a 73-frame render takes roughly 4 to 6 minutes. When the generation finishes, the Save Video node outputs your finished MP4 file with the complete 360-degree rotation. Tips Subject framing: Keep your subject centered in the frame with some room around the edges. If a person’s elbow or hair is clipped by the image boundary, the model may struggle to generate a circular motion. Depth contrast: Images with clear separation between the subject and the background produce the strongest effect.. 73 frames: MiniMax H3 processes video latents on a strict 17k + 5 frame boundary (22, 39, 56, 73, 90 …) Deviating from these magic frame numbers can cause the camera to stutter. Action shots: In my testing, action pictures featuring airborne elements look the most impressive in frozen-time orbits. What’s next MiniMax H3 Image to Video MiniMax H3 Text to Video MiniMax H3 Fun ControlNet Union Wan 2.2 Image to Video AnimateDiff ComfyUI The post Turn Any Still Photo Into a 360 Orbit Infinite Loop Video first appeared on Stable Diffusion Art.