MiniMax H3 reference-to-video with 8 photos: the $0.10 extra per image
On MiniMax H3, reference images beyond the first five add $0.10 each on Sume. Eight photos at 10 s, 768p cost $1.05; five cost $0.75.

On MiniMax H3, a reference-to-video request includes five reference images in the per-second price, and every image after the fifth adds $0.10 on Sume. A 10-second clip at 768p costs $0.75 with five photos and $1.05 with eight. The extra is the vendor list add-on of $0.08 per image times the Sume multiplier of 1.25.
H3 Max follows a different rule in Sume: it bills the per-second rate with no per-image add-on.
The arithmetic for a 10-second 768p clip
H3 lists $0.06 per second at 768p. Ten seconds is $0.60. Each reference image after the first five lists at $0.08. Sume multiplies the total by 1.25 and rounds up to the cent.
| Reference images | Extra images | List total | Sume price |
|---|---|---|---|
| 1 to 5 | 0 | $0.60 | $0.75 |
| 6 | 1 | $0.68 | $0.85 |
| 7 | 2 | $0.76 | $0.95 |
| 8 | 3 | $0.84 | $1.05 |
| 9 | 4 | $0.92 | $1.15 |
H3 versus H3 Max for many references
H3 Max uses the same reference-to-video envelope as H3, and lists $0.08 per second at 768p. A 10-second 768p H3 Max clip is $1.00 in Sume before any such overage, whatever the photo count. H3 costs less for up to five images: $0.75 versus $1.00. At eight images the H3 price of $1.05 is slightly above the H3 Max base price.
There is a catch on the H3 Max side. A related Sume post covers reference-token overage on H3 Max, which is not part of the reserve. If your clip needs exact cost control, price the job with the vendor page open and compare it with the usage.cost in the finished job.
How to keep the photo count down
- Use five images: one subject, one product, one setting, one style board, one detail.
- Merge related items into a single collage photo instead of sending three separate packshots.
- Drop duplicates, and remember the model takes at most nine images. Two near-identical angles of the same object add cost and confuse the model.
- Test with three photos at 480p first, then add references only if the clip is missing something.
A catalogue budget
A shop that renders 50 product clips of 10 seconds at 768p, each with eight reference photos, pays 50 x $1.05 = $52.50. With five photos each the figure is 50 x $0.75 = $37.50, so the three extra photos per clip add $15.00. That is a 40 percent increase for photos the model may not need.
The same 50 clips on H3 at 480p with five photos cost 50 x $0.63 = $31.50, because 10 seconds at $0.05 lists $0.50 and Sume rounds $0.625 up to $0.63. Run the draft at 480p first and spend on the extra references only for the clips you keep.
Send the references
The model takes up to nine reference images, and they go in input_references. They guide the content of the clip, and no image is forced to be the opening frame.
import os
import time
import requests
API = "https://api.sume.com/v1/videos"
HEADERS = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
body = {
"model": "minimax-h3",
"prompt": "The red sneaker rotates on a concrete plinth in the studio from the photos, soft top light",
"duration": 10,
"resolution": "768p",
"input_references": [
{"type": "image_url", "image_url": {"url": "https://example.com/shoe-1.jpg"}},
{"type": "image_url", "image_url": {"url": "https://example.com/shoe-2.jpg"}},
],
}
job = requests.post(API, headers=HEADERS, json=body, timeout=60)
job.raise_for_status()
poll_url = job.json()["polling_url"]
while True:
state = requests.get(poll_url, headers=HEADERS, timeout=60).json()
if state["status"] in ("completed", "failed", "cancelled"):
break
time.sleep(10)
print(state["status"], state.get("usage"), state.get("unsigned_urls"))
What to verify yourself
The fal H3 reference-to-video page says the first five reference images are free, each additional image costs $0.08, and the model takes up to nine images. Vendors change pricing, so compare the fal H3 page with the Sume catalog before a large batch. Nine images for 10 seconds at 768p is $1.15 on Sume.
Sources
Related posts
More in Pricing
- Nano Banana 2.1 batch: $0.0168 per 1K image, 100,000 images cost
Google's batch price for Nano Banana 2.1 is $0.0168 per 1K, $0.0252 per 2K and $0.0567 per 4K image: half of Standard. What 100,000 images cost in each tier.
- Nano Banana 2.1 costs $0.0336 at Google and $0.10 on Sume: the gap
Google lists Nano Banana 2.1 at $0.0336 for 1K; Sume's card is $0.10. Direct is cheaper per image. Here is the math, and when the Sume route still fits.
- Nano Banana 2.1: 100 drafts at 0.5K plus 10 finals at 4K is $9.50
On Sume, Nano Banana 2.1 bills $0.075 at 0.5K and $0.20 at 4K. A 100-draft, 10-final run costs $9.50 against $20.00 for 100 images at 4K.
- Nano Banana 2.1 prices: Google's 1K and 4K list vs Sume's cost
Google launched Nano Banana 2.1 on Oct 6 at $0.0336 per 1K image and $0.0756 at 4K. Here is how to read what Sume bills for google/nano-banana-2.1.
Written by Sume