Reference to video AI: keep a character or product consistent across a clip

Makernice reference to video generates a clip from a prompt plus several reference images, so a character, product or style stays consistent without fixing any frame. It is one mode of the Makernice AI video generator: 9 of its 13 video models support it, with up to 16 images. Every run is quoted first and billed per second.

Makernice Team · Updated November 2, 2026

How do you make a reference-to-video clip on Makernice?

Use the web console for a few clips without code, or the API for apps and batches. Both use the same models, the same prices and the same prepaid balance.

In the browser (no code)

Sign in and top up your prepaid balance (from $15; it never expires).
Open Create → Video, pick a model that offers Reference, then upload your reference images (JPEG, PNG or WebP, up to 10 MB each) and describe the shot.
The price appears before you press Generate. Download the clip when it's ready; finished files are kept for about 30 days.

Open the video console

With the API

Quote: POST /v1/generations/quote returns the exact amount, holds nothing and locks the price for 12 minutes.
Create: POST /v1/generations with the quote_id, the same input and an Idempotency-Key header. The quoted amount is held from your balance.
Poll GET /v1/generations/{id} until status is succeeded, or add a top-level callback_url to get a webhook. Then download outputs[0].url.

Quickstart · Webhooks · No charge for failed generations

curl https://api.makernice.com/v1/generations/quote \
  -H "Authorization: Bearer $MAKERNICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"kling-4.0","mode":"r2v","prompt":"The woman from the reference photos walks through a night market holding the blue ceramic mug, handheld camera","reference_images":["https://example.com/woman-front.jpg","https://example.com/woman-side.jpg","https://example.com/blue-mug.png"],"duration":5,"resolution":"720p","aspect_ratio":"16:9"}'

curl https://api.makernice.com/v1/generations \
  -H "Authorization: Bearer $MAKERNICE_API_KEY" \
  -H "Idempotency-Key: 7c1e9a2b-order-1842" \
  -H "Content-Type: application/json" \
  -d '{"quote_id":"qt_…","input":{"model":"kling-4.0","mode":"r2v","prompt":"The woman from the reference photos walks through a night market holding the blue ceramic mug, handheld camera","reference_images":["https://example.com/woman-front.jpg","https://example.com/woman-side.jpg","https://example.com/blue-mug.png"],"duration":5,"resolution":"720p","aspect_ratio":"16:9"}}'

curl https://api.makernice.com/v1/generations/gen_… \
  -H "Authorization: Bearer $MAKERNICE_API_KEY"

Which Makernice video models support reference to video, and how many images can each take?

9 of the 13 video models accept mode r2v; MiniMax H3 Max Turbo, Kling 3.0 Pro, Kling 3.0 Standard and Kling 3.0 4K do not. Seedance 2.5 takes up to 16 images, and Wan 3.0, Wan 3.0 Prime and Kling 4.0 take up to 10. Wan 3.0, Wan 3.0 Prime and Happy Horse 1.1 do not accept aspect_ratio in this mode.

Catalog facts from GET /v1/models (maxReferences per mode). MiniMax H3: 480p, 768p native · 2K, 4K upscaled from 768p. MiniMax H3 Max: 480p, 768p, 1080p (refined from 768p). Request errors and how to pick images: Reference-to-video guide.
Model	Max reference images	Aspect ratios in r2v	Duration	Resolutions
Seedance 2.5	16	adaptive · 16:9 · 9:16 · 1:1 · 4:3 · 3:4 · 21:9 · 9:21	4–30 s	480p, 720p, 1080p, 4K
Seedance 2.0	9	16:9 · 9:16 · 4:3 · 3:4	4–15 s	720p, 1080p
Seedance 2.0 Fast	9	16:9 · 9:16 · 4:3 · 3:4	4–15 s	720p (fixed)
MiniMax H3	5	21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16	5–15 s	480p, 768p, 2K, 4K
MiniMax H3 Max	4	adaptive · 21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16	5–15 s	480p, 768p, 1080p
Wan 3.0	10	Not accepted	2–30 s	480p, 720p, 1080p
Wan 3.0 Prime	10	Not accepted	2–30 s	480p, 720p, 1080p
Kling 4.0	10	16:9 · 9:16 · 1:1 · 21:9	3–30 s	720p, 1080p
Happy Horse 1.1	9	Not accepted	3–15 s	720p, 1080p

Catalog facts from GET /v1/models (maxReferences per mode). MiniMax H3: 480p, 768p native · 2K, 4K upscaled from 768p. MiniMax H3 Max: 480p, 768p, 1080p (refined from 768p). Request errors and how to pick images: Reference-to-video guide.

How much does a 10-second reference-to-video clip cost on Makernice?

At each model's default resolution, a 10-second reference-to-video clip costs from $0.69 on MiniMax H3 Max to $2.59 on Seedance 2.5 (price list v12, November 3, 2026). On every model in this table, reference to video costs the same per second as text to video at the same resolution. The number of reference images does not change the price.

From GET /v1/pricing, at each model's default resolution. Totals are rounded to the cent; your quote shows the exact amount before you run anything. Live rates: Pricing. MiniMax H3: 480p, 768p native · 2K, 4K upscaled from 768p.
Model	Resolution	Per second	5 s clip	10 s clip
Seedance 2.5	720p	$0.2592/s	$1.30	$2.59
Seedance 2.0	720p	$0.1704/s	$0.85	$1.70
Seedance 2.0 Fast	720p	$0.1027/s	$0.51	$1.03
MiniMax H3	2K	$0.156/s	$0.78	$1.56
MiniMax H3 Max	768p	$0.0685/s	$0.34	$0.69
Wan 3.0	720p	$0.0789/s	$0.39	$0.79
Wan 3.0 Prime	720p	$0.1234/s	$0.62	$1.23
Kling 4.0	720p	$0.168/s	$0.84	$1.68
Happy Horse 1.1	720p	$0.168/s	$0.84	$1.68

From GET /v1/pricing, at each model's default resolution. Totals are rounded to the cent; your quote shows the exact amount before you run anything. Live rates: Pricing. MiniMax H3: 480p, 768p native · 2K, 4K upscaled from 768p.

Prices as of November 3, 2026 (price list v12).

What should you check before sending a Makernice reference-to-video request?

Put every image in reference_images as a public https URL of up to 2,048 characters, at least one and at most the model's limit. Do not send image_url or last_image_url in this mode.
More than 10 images only works on Seedance 2.5; switching models with a long image list needs a shorter list.
Send aspect_ratio only to models that list it; Wan 3.0, Wan 3.0 Prime and Happy Horse 1.1 take none in this mode.
Only upload images you own or have permission to use, and don't upload photos of other people without their consent.

Frequently asked questions

What is the difference between reference to video and image to video on Makernice?

Image to video uses one image as the first frame of the clip. Reference to video (mode r2v) takes several images in reference_images, up to 16 on Seedance 2.5, to keep a character, product or style consistent without fixing any frame. 9 of the 13 video models support it.

Do more reference images cost more on Makernice?

No. A reference-to-video clip is priced by the model, the resolution and the seconds you request. Sending one reference image or the model's maximum gives the same quote.

Which Makernice video models cannot do reference to video?

MiniMax H3 Max Turbo, Kling 3.0 Pro, Kling 3.0 Standard and Kling 3.0 4K. MiniMax H3 Max Turbo, Kling 3.0 Pro and Kling 3.0 Standard still do image to video and first and last frame; Kling 3.0 4K does text to video only. For several reference images, pick one of the 9 models that support reference to video.

Does Makernice charge for a failed reference-to-video generation?

No. Creating a run holds the quoted amount. If the generation fails, times out or is canceled while queued, the hold is released in full. Quality is not a failure condition: a delivered video is charged, whatever you think of it. Note: if a finished output is blocked by output moderation for a policy violation, the run is still charged. You can contact support for a manual review.

AI video generator · Text to video · Image to video · Reference-to-video guide · Pricing