MiniMax H3 Text-to-Video: Text to Video Online

Every row comes from VdoBloom's live model configuration, the generation endpoint and the credit table. The detail people most often miss: this mode takes no image at all. If you want to start from a photo, the sibling page, MiniMax H3 Image-to-Video, is the one you need; it shares the price list but not the input.

The price is linear: seconds times 4 at 768P, seconds times 6 at 2K, with no minimum fee on top. The dollar figures use the Lite plan rate of $15 for 300 credits ($0.05 a credit). New accounts get 10 free credits, which does not reach the 16-credit minimum for this model, so plan on a top-up before your first H3 render.

MiniMax H3 appears in the picker as two separate variants, and they are not interchangeable. The price list and the 4 to 15 second range are identical; what differs is what you hand the model and what you control afterwards. Text-to-video gives you the frame shape and leaves the look to the prompt. Image-to-video gives you the exact first frame and takes the frame shape from your photo.

All five accept a text prompt in the Text to Video tab, so switching is a dropdown. The plain summary from the credit table: H3 offers any-second lengths up to 15 seconds and a 2K tier, and its 768P rate (4 credits a second) sits below Seedance 2 at 720p (82 credits for 5 seconds), Seedance 2.5 at 720p (25.2 a second) and Kling 3.0 standard (5.6 a second, more with sound). Kling 3.0 offers three ratios, with sound as a paid option. Hailuo 02 only renders 6 or 10 seconds and costs more per clip than H3 at both lengths. Seedance 2.5 is the pick when one take must run past 15 seconds. 768P and 720p are not identical resolutions, so read the middle columns as 720p-class.

Frame shape is the one thing this mode controls that the image-to-video sibling cannot, so it is worth choosing on purpose rather than accepting the 16:9 default. The ratio does not change the price: a 5-second 768P clip is 20 credits in every shape.

No tab opens on this model by default, so you select it. The model, the 4 to 15 second range and the per-second rate are the same everywhere it appears.

With no image to anchor it, the prompt carries the whole shot, and you have up to 7,000 characters for it. Open with the subject and the action in one sentence, because that is what the clip has to deliver. Then add the camera (a slow push-in, a handheld follow, a locked-off wide), the setting and time of day, and the light. Describe the sound you want too: rain on a window, a crowd murmur, footsteps on gravel. The audio is generated with the picture, so a scene with no sound cue in the prompt leaves that choice to the model.

For longer clips, write the shot in beats (what happens in the first few seconds, then the middle, then the end) instead of one line, which tends to stall after the first moment. Match the composition to the ratio: a 9:16 prompt about a sweeping horizon fights its frame. Draft at 768P and 4 to 5 seconds (16 to 20 credits), then re-run only the winning prompt at 2K or a longer length. Because lengths come in one-second steps, you can render the exact number of seconds your edit needs instead of trimming a longer clip you paid for.

This mode cannot take an image, a last frame, reference photos or a reference video. For a specific product, logo or face, use MiniMax H3 Image-to-Video, or a model with reference inputs such as Seedance 2.5. The ceiling is 15 seconds per clip; longer single takes need Seedance 2.5. 2K costs 50% more per second than 768P (6 against 4 credits), so drafting at 2K is the most common way to overspend on this model. The 10 free signup credits do not cover the 16-credit minimum. H3 is a proprietary MiniMax model that VdoBloom runs through its provider, KIE; there are no downloadable weights, and if a job fails on the provider side the credits come back automatically.

VdoBloom does not rank models on subjective quality. Everything above is spec and price; whether H3's look suits your shot is something a 16-credit 4-second draft answers faster than any review.