The reviewed local configuration
For T2V and I2V, the current environment selects a pruned W4A8 FL2VA checkpoint. It also selects a 4B encoder with a matching projection file. The R2V path uses a separate reference checkpoint. These distinctions matter when investigating compatibility or comparing results from different installations.
Smaller representations do not justify a blanket claim of unchanged quality or a fixed speed improvement. The complete graph, offloading behaviour and requested clip size influence the outcome. A file's disk size is not a direct measurement of the total GPU memory needed to generate a clip.
Evaluate a deployment change
- Save the complete working configuration and a small representative prompt set.
- Include text, image and reference cases if those workflows are affected.
- Keep request settings fixed while comparing the old and new configuration.
- Record failures, processing time and visual differences, not only whether the model loads.
What hosted users should do
Choose the correct workflow and use a clear source rather than trying to specify a quantization format in the prompt. If you need a reproducible production asset, retain the output and its request details; a later service update can change results even when the creative text stays the same.
Keep claims specific
Describe a measured file or a reviewed implementation, not “full quality at any resolution.” Our guides identify local settings where verified and avoid treating the official maximum capability as a promise for every JollyAI preset. Technical snapshots should be updated when the serving configuration changes, especially after replacing an encoder or projection.
REVIEWED JOLLYAI IMPLEMENTATION
The workflow behind the guide
Written scene → text conditioning → turbo audio/video sampling → separate video and audio decoding → combined video file.
- Inputs, presets & access
- Limited free access; four-second free preset, with six/eight/ten-second PRO options. Current ordinary pixel budget: 0.4 MP. Use the account controls for current availability.
- Implementation note
- Local H3 turbo workflow. Six sampling steps; simple scheduler; turbo adapter; video/audio shifts 12/3. Current environment: pruned W4A8 FL2VA weights with a 4B encoder and matching projection. R2V uses a separate reference checkpoint. These settings describe the reviewed JollyAI deployment, not every H3 implementation.
Configuration reviewed September 20, 2026. Your current playground controls and plans page determine availability. Prompt recipes above are suggestions, not guarantees or claimed test results.
Open Text to video ↗Sources and useful next steps
The implementation notes describe JollyAI's reviewed local integration. Upstream capabilities and licences may differ from the settings available here.
- MiniMax H3 upstream project and model information ↗
- MiniMax H3 tool overview on JollyAI
- Earlier H3 ComfyUI workflow snapshot and downloadable JSON — check the newer configuration notes above when reproducing it.
- JollyAI content policy and service terms
Found an outdated setting or a reproducible problem? Contact JollyAI with the workflow and job reference. Keep account tokens and private input files out of public reports.