Video Generation

Create a generated video

Add a video-generation item from the add menu on a video track. Shrimply loads the models and controls offered by the selected compute server. Choose a model, provide the requested text and media inputs, then start generation.

Generation continues through the compute server, so keep the server running until the result has returned to the editor. Cancel the operation from Shrimply if you no longer need it.

Available models

The current server supports these model families:

MiniMax H3 Base

Text-to-video, first/last-frame-to-video, and ordered reference-to-video.

Looping Sketch Anime

Text-to-video and first/last-frame-to-video using the pinned animation adapter.

Wan 2.1 T2V 1.3B

Text-to-video in landscape or portrait orientation.

Wan 2.2 TI2V 5B

Text-to-video and first-frame image-to-video in landscape or portrait orientation.

The editor uses the server’s live model catalog. If a model is not listed, it is not currently available from the selected server.

Before downloading

Video-generation models can require substantial download space, memory, and processing time. Requirements vary by model and by the selected compute device. Check the model publisher’s documentation before downloading weights, especially on a shared or storage-constrained machine.

MiniMax H3 weights are governed by the separate MiniMax H3 Community License Agreement. Review and accept it before using the model.

Wan model details and licenses are available from the official Wan 2.1 1.3B and Wan 2.2 5B pages.

Memory controls

Some models offer Normal and Low VRAM modes. Low VRAM reduces accelerator-memory pressure by moving more work through system memory and is usually slower. Quantization can reduce model memory further, with a possible quality or performance tradeoff.

Start with the default settings. Use the reduced-memory options only when the default cannot run on the selected server.