nano-banana.live has no video-generation API and no upload endpoint for this tool. The extractor below uses a video element and a canvas in your browser. The file is not sent to this site. Turning the two images into a transition still requires a model or a hosted service that accepts a first frame, a last frame, or both.
Extract the first and last frame
Choose a video file. Nothing is uploaded.
What the two frames are for
First-and-last-frame generation is a conditioning pattern. You supply the opening image, the closing image, or just one of them, and a video model fills the frames between those constraints. The images are inputs. They are not the finished video, and saving them does not start a render.
One published implementation is MiniMax H3-Base FL2VA. Its model card, re-read October 22, 2026, says that checkpoint accepts zero, one, or two images: none is text-to-video, one is a first-frame or last-frame condition, and two are first-and-last-frame generation. MiniMax's local-deploy guide, last reviewed August 26, 2026, maps that to the ComfyUI I2V template and to SGLang task fl2va. Hugging Face's Diffusers documentation for the same checkpoint names the inputs image and last_image. Those pages are the workflows. This website does not call them.
Use the frames in a real workflow
- Pick a video you have the rights to use. The status line should keep saying the file stays in this tab.
- Wait until the status line reports both a first-frame time and a last-frame time, then download the two PNGs. The pixel size is whatever the browser decoded, not a model's required canvas.
- Treat the PNGs as conditions. A generator still has to produce the in-between frames, the frame rate, and, for H3-Base, the stereo audio.
- For a documented local path, follow How to Run MiniMax H3 Locally. That guide separates H3-Base weights from MiniMax's hosted API. H3-Context-IR and H3-Regenerate-2K are not in the open weights.
- If you want MiniMax to run the job on their servers, use the platform linked from the model card. A script on your computer that only calls that API is still cloud inference. Image-only local routes on this site, such as the local AI image generator guide, do not perform this video task.
FAQ
Does this page generate a first-and-last-frame video?
No. The tool on this page only reads a video file in your browser and saves the opening frame and the closing frame as PNG images. It does not call a video model, and it does not upload the file to nano-banana.live.
Are videos uploaded to a server?
No. The file is opened with a local object URL, drawn to a canvas in this browser tab, and downloaded from that canvas. Closing the tab or choosing another file drops that local copy. This page has no generate request and no video upload endpoint.
What is first-and-last-frame video generation?
A generator is given the image a clip should start on, the image it should end on, or both, and it produces the frames in between. MiniMax's H3-Base FL2VA model card describes that input pattern: no image is text-to-video, one image is a first-frame or last-frame condition, and two images are first-and-last-frame generation. Checked October 22, 2026.
Can nano-banana.live run that generation for me?
No. This site does not host MiniMax H3 and does not proxy the MiniMax API. The local-run guide links the documented H3-Base paths. MiniMax's own platform is the hosted service. A script on your computer that only calls their API is still cloud inference.
Why can the saved last frame be slightly before the final timestamp?
Browsers often cannot seek to the exact duration. This tool seeks a short step before the reported duration and draws the frame the browser decodes there. On a one-frame or very short file, the first and last images can match.