Comparisons
Compare MiniMax H3 and H3 LoRA by resolution, video-reference restrictions, aspect ratios and complete input-plus-output costs.
MiniMax H3 and H3 LoRA both support reference-driven video generation. Choose by output resolution, video input and framing first, then calculate the complete request cost. The LoRA name establishes neither a quality winner nor the cheaper total.
Both currently offer text-to-video, image-to-video and reference generation, with 5–15-second output and limits of 9 images, 3 videos and 3 audio clips. Reference generation is not exclusive to LoRA. H3 model page, H3 LoRA model page
The LoRA public contract does not accept custom LoRA configuration. Neither page establishes a custom-weight training workflow.
Current differences include:
Of these two routes, H3 currently meets a 2K requirement. LoRA offers direct 480p output or 9:21 framing. A 768p project with video input must account for LoRA's separate video-reference restriction.
The public APIXO Pricing entries checked on October 10, 2026 list 768p output at $0.09 per second for H3 and $0.08 for LoRA. LoRA also charges $0.02 per reference image and $0.02 per reference audio clip.
API usage rates; Studio monthly credits are separate.
For a hypothetical 5-second 768p output with 5 reference images, 2 audio clips and no video reference:
These are calculations, not recorded charges. The input combination reverses the apparent price order. Check APIXO Pricing first for current rates; the relevant entries were checked through public search and match the model-page rates.
Video references add their actual input seconds. If LoRA uses 480p while H3 uses 768p, that comparison no longer has matching output specifications.
The documented gaps are specific: direct 480p output and 9:21 framing are listed on LoRA rather than the current H3 route.
There is insufficient evidence here to claim that only LoRA can achieve a particular character, style or action. Similarly, 2K resolution does not guarantee correct product geometry or complete motion.
Check non-negotiable delivery requirements first: 2K or video references at 768p, versus 480p or a 9:21 ratio. Eliminate routes that do not meet them before comparing rates. Then price the full reference set.
For identical 768p settings without video references, let d be output seconds, n reference images and a reference audio files. The verified current Pricing entries give:
Five seconds with no image/audio references costs $0.45 for H3 and $0.40 for LoRA. Adding five images and two audio files changes that to $0.45 versus $0.54. These are calculated examples for the stated comparable settings, not submitted tasks.
What does LoRA cover that ordinary H3 currently does not? Of these two pages, LoRA lists 480p and 9:21, which H3 does not currently list. Ordinary H3 already has multimodal references, so reference-image generation is not a LoRA-only capability. Differences in a particular style, character or action require comparable output samples.
For project comparisons, divide actual spend by accepted clips or seconds. Keep rejected outputs and repeats in the budget rather than removing them from the record.
Use the same images and motion brief under shared supported settings. Retain output cost, input cost, failed attempts and reasons for rejecting clips.
Inspect shape, identity, action and sound against the business requirements. A more detailed image requiring extensive corrections may be less useful. No generated samples were collected for this guide, so there is no measured speed, success-rate or quality ranking.
First eliminate incompatible specifications, then calculate the full input combination, then test a small set. Evaluate H3 LoRA for suitable low-resolution drafts and H3 for 2K or its supported video-reference settings.
Build with APIXO
Explore image, video, and audio models in one place.