One-still I2V credit floors

One still in, clip out. The Versely floor is min_credits on the row you dispatch, read from models.json on 19 September 2026. It is not a quality ranking. It is not a per-second cell you multiplied by a length the picker will not sell. Wan 3.0 Image to Video sits at 6. VEO 3.1 Fast sits at 16. Kling 3 Turbo sits at 22. Seedance 2.5 sits at 34. Kling Video V3 Standard Image to Video sits at 51. VEO 3.1 sits at 64. Some of those rows have a public model page. Some do not. This page quotes both and only links published slugs.

The job is one jpeg (or png) and a motion pass. If you have no still, you are not on this page. If you have a stack of references, you wanted reference-to-video. If you have a first frame and a last frame, you wanted that pair. One still is the whole input.

The floor is the shortest clip you can buy

A floor is the cheapest legal generate on that row. It is min_credits at the shortest duration the row lists, on the cheapest band the matrix sells, after any catalog discount already baked into that integer. It is not "credits per second times the length you had in mind." The picker will not sell a two-second Seedance take. It will not prorate Kling Video V3 Standard's first five seconds down to three. It will not turn Turbo's silent plate into a talking clip because a vendor blog said audio ships at source.

Image-to-video is a flag, not a vibe. On the catalog, requires_image: true means the job will not start without the file. Several families also ship a text-to-video sibling at the same rate card. This page uses the sibling only when the I2V slug is unpublished, and it says so. Do not paste a still into a text-to-video row and assume you paid the I2V meter. Do not skip the still on a row that requires it.

The generate door is the AI video generator. Filter by the input you actually have in the Model Finder. Neither tool changes the floor. They only stop you from opening the wrong family.

Crop and ratio before you spend. I2V inherits the still. A 1:1 product plate will not become a 9:16 hook because you typed "vertical" in the prompt. A letterboxed screenshot will letterbox the clip. The floor buys motion on the pixels you uploaded. It does not buy a recrop.

What the floor is not:

• A USD price. Versely credits are credits. This page does not convert them.
• A quality score. A 6-credit floor and a 64-credit floor are different meters, not a ranking of "which model is better."
• A 15-second quote. max_credits is the other end of the published range. Read it when the take is long. Do not budget the floor for a 30-second pass.
• A promise that every slug below has a /models/ URL. Unpublished rows are quoted as catalog facts. They are not linked.

Six 19 September rows, quoted from models.json

All numbers below are from src/data/models.json on 19 September 2026. Where a public page exists, the link is that page. Where it does not, the slug is named in text and not turned into a URL.

Wan 3.0 Image to Video (catalog only, not a published model page). min_credits 6, max_credits 168, full_min_credits 8, full_max_credits 240. Audio true. requires_image true. Category image-to-video. Durations 2s through 15s. 30 percent off. Resolution-based per second: SD 4 full / 3 discounted, HD 8 / 6, 4K 16 / 12. Two seconds at the SD discounted cell is 2 × 3 = 6, which matches the floor. That is why this row can undercut every other card on this page: the picker sells a two-second take, and SD is cheap. Fifteen seconds at the 4K discounted cell would be 15 × 12 = 180. The listed max is 168. Thirty percent off 16 × 15 is 168. Report both. The live quote is the bill. Do not link /models/wan-3-0-image-to-video. It is not in scripts/_published-models.json.

VEO 3.1 Fast (catalog only). min_credits 16, max_credits 48, full_min_credits 32, full_max_credits 96. Audio true. Category image-to-video. requires_image is false on the card even though the category is I2V: the still is the job you came for, not a hard flag on this row. Durations 4s, 6s, 8s. 50 percent off. Audio off 8 full / 4 discounted per second. Audio on 12 / 6. Four seconds, audio off, discounted: 4 × 4 = 16, which matches the floor. Eight seconds, audio on, discounted: 8 × 6 = 48, which matches the max. Fast is the cheap Veo-shaped meter for a still. It is not the published Veo page. Do not link /models/veo-3-1-fast.

Veo's public surface. The Versely floor for the published VEO 3.1 row is 64 credits. Fast at 16 is a separate catalog card without a published model URL.

Kling 3 Turbo, same rate on both catalog rows. Kling 3 Turbo Text to Video is the published door: min_credits 22, max_credits 108, full_min_credits 27, full_max_credits 135. Audio false. Durations 3s through 15s. 1080p ceiling. 20 percent off. Per second: 9 full, 8 discounted. The image-to-video sibling in models.json uses that same card (requires_image true, audio false, 22 to 108). That sibling is not published. Do not link /models/kling-3-turbo-image-to-video. Link the T2V page and bring the still to the I2V row in the studio. Three seconds times 8 is 24. The listed min is 22. That is a 2-credit gap. Twenty percent off 9 is 7.2. Three times 7.2 is 21.6, which rounds to 22. Fifteen times 7.2 is 108, which matches max_credits. Fifteen times 8 is 120, which does not. Report the cell (8) and the floor (22). Plan silent video. Captions and voice are a later meter.

Kling's public home. Turbo on Versely is audio off. The 22-credit floor is a silent 3-second plate, not bundled dialogue.

Seedance 2.5. Seedance 2.5 is the published family row: min_credits 34, max_credits 557, full_min_credits 67, full_max_credits 1114. Audio true. Durations 4s through 30s. 720p ceiling. 50 percent off. Resolution-based per second: SD 17 full / 9 discounted, HD/4K 38 / 19. Category on that published slug is text-to-video, requires_image false. The image-to-video sibling seedance-2-5-image-to-video copies the same 34 / 557 / audio true / 4-30s card and sets requires_image true. That sibling is not published, so it has no /models/ link here. One still on Seedance is that sibling, at the same floor. Four seconds times the SD discounted cell of 9 is 36. The listed min is 34. That is a 2-credit gap. Half of 17 is 8.5. Four times 8.5 is 34, which matches min_credits. Thirty seconds times the HD/4K discounted cell of 19 is 570. The listed max is 557 (exactly half of full_max_credits 1114). Report both. The 34 floor is a 4-second SD take, not a 30-second hero.

Kling Video V3 Standard Image to Video. This is the published I2V row in the set that actually requires the still. Kling Video V3 Standard Image to Video: min_credits 51, max_credits 152, full min and full max the same (no discount). Audio true. requires_image true. Durations 3s through 15s. Billing is base_plus_per_second: first 5s 51, each extra second 11. Three seconds is still 51. Four seconds is still 51. Five seconds is 51. There is no prorate inside the opening block. Ten seconds is 51 + 5 × 11 = 106. Fifteen seconds at that matrix would be 51 + 10 × 11 = 161. The listed max is 152. Those two disagree by more than a rounding tick. This page will not invent a third extra-second rate to force 152. Confirm the 15-second quote in the app before you dispatch a long take. Confirm the 3-second quote too: you will pay 51 either way, so take the five seconds you already bought and trim later if the platform cap is shorter.

VEO 3.1. VEO 3.1 is the published Veo row: min_credits 64, max_credits 256, no discount. Audio true. Category text-to-video, requires_image false. Durations 4s, 6s, 8s. 4K ceiling. Audio-based per second: audio off 16, audio on 32. Four seconds, audio off: 4 × 16 = 64, which is the floor. Four seconds, audio on: 4 × 32 = 128. Eight seconds, audio on: 8 × 32 = 256, which is the max. If the still job needs native sound on this row, 64 is the wrong budget. 64 is the silent 4-second end of the published Veo card. Fast at 16 is the cheaper Veo-shaped I2V card, unpublished, quoted above.

Row	Published /models/?	min	max	Audio	Shortest duration	Still required on the card
Wan 3.0 Image to Video	No. Cite only	6	168	true	2s	true
VEO 3.1 Fast	No. Cite only	16	48	true	4s	false (category is I2V)
Kling 3 Turbo (T2V published, I2V same rates)	T2V yes. I2V no	22	108	false	3s	I2V sibling true
Seedance 2.5 (T2V published, I2V same rates)	T2V yes. I2V no	34	557	true	4s	I2V sibling true
Kling Video V3 Standard I2V	Yes	51	152	true	3s (bills as first 5s)	true
VEO 3.1	Yes	64	256	true	4s	false

That table is the page. Rank by the floor if the question is "what does one still cost to start." Do not rank by a review score you did not measure today.

When min credits and per-second math disagree

Catalog rows store two stories that are allowed to be one or two credits apart.

The matrix stores a per-second (or per-block) cell, often already rounded. min_credits stores the integer the estimator will not go below. Discounts are stored as percents (20, 30, 50) and as those rounded cells. Half of an odd number is not an integer. Twenty percent of 9 is 7.2, not 8. If you multiply the rounded cell by the shortest duration, you can land one or two credits above the listed floor.

On this 19 September snapshot, the honest gaps are:

• Seedance 2.5. SD discounted cell 9 × 4s = 36. Listed min 34. Gap of 2. Half of full SD 17 is 8.5, and 4 × 8.5 = 34, which matches the floor. Full min is 67 against 4 × 17 = 68 (gap of 1).
• Kling 3 Turbo. Discounted cell 8 × 3s = 24. Listed min 22. Gap of 2. 20 percent off 9 is 7.2, and 3 × 7.2 = 21.6, which rounds to 22. The same 7.2 × 15 = 108, which matches max. The cell × 15 = 120, which does not.
• Wan 3.0 Image to Video. 2s × SD discounted 3 = 6, which matches min. The long end disagrees if you use the 4K discounted cell of 12 (15 × 12 = 180) against listed max 168. 30 percent off 16 × 15 = 168. Report both.
• Kling Video V3 Standard I2V. The short end does not have a 1-2 credit gap. 3s, 4s, and 5s are all 51 because the first block is indivisible. The long end (matrix 161 vs listed max 152) is a larger disagreement. Do not paper over it. Read the quote.
• VEO 3.1 and VEO 3.1 Fast. Short-end arithmetic matches the listed mins (64 and 16) when you use the matching audio switch.

When the cell and the min disagree by a credit or two, this page prints both. It does not pick a favorite. The app shows a whole-credit total before you confirm. That total is the charge. A blog table is a snapshot of the catalog file, not a second meter.

Do not build a personal "true rate" and reuse it next month. Discounts move. Cells get re-rounded. min_credits is the field to grep. If it moved, this table is stale and the live row wins.

What a floor is not

A floor is not a 30-second budget. Seedance's 34 is the 4-second SD end of a card that also lists 557. If you need the long hold, you are no longer talking about a floor. You are talking about duration times a band. Same still. Different bill.

A floor is not native audio unless the row says so. Turbo's 22 is silent (audio: false on both catalog rows). Seedance, Wan I2V, Kling V3 Standard I2V, Veo 3.1, and Veo Fast list audio true. True means the model is allowed to emit sound. It does not mean every 4-second take includes the dialogue you imagined. It does not mean Turbo will start talking if you prompt "voiceover." For Turbo, plan burned captions and a voice pass. For Veo 3.1, audio on doubles the per-second cell from 16 to 32, so the 64 floor is specifically audio off.

A floor is not a published URL. Wan I2V at 6, Veo Fast at 16, Turbo I2V, and Seedance I2V are real rows in models.json and they are not in _published-models.json as of this snapshot. Linking a /models/ slug that the site does not publish is a 404 with extra steps. Cite the number. Use the studio picker. Link the published sibling when the rate card is the same.

A floor is not first-last, motion control, lipsync, or a one-tap template. Those jobs take different inputs (two stills, a driving clip, a face plus audio, a template id) and different meters. One still, clip out, is I2V. Keep it boring on purpose.

A floor is not five rerolls. Each dispatch bills. The 6-credit Wan take you reject still spent 6. The 51-credit Kling V3 Standard take you trim to three seconds still spent 51. Budget the tries you will actually run.

If the still is already on disk, create a Versely account and generate from the on-screen quote. Use this table to pick the row, not to override the quote.

FAQ

Why is Wan 6 and Veo 64 for the same still?

Different shortest clips, different bands, different discounts. Wan I2V sells 2 seconds at an SD discounted 3 credits a second (2 × 3 = 6) and is unpublished as a model page. VEO 3.1's published floor is 4 seconds with audio off at 16 a second (4 × 16 = 64). They are not comparable as "quality per credit." They are two legal starts. Pick the row for the length, audio flag, and whether you need a published Veo door or a cheap catalog I2V meter.

Can I open unpublished Turbo I2V or Veo Fast model pages?

Not from this site, not today. kling-3-turbo-image-to-video and veo-3-1-fast exist in models.json and they do not exist in scripts/_published-models.json on 19 September 2026. Turbo I2V matches the published T2V rate card (22 min, 108 max, audio false, 8 discounted on the cell). Fast is 16 min, 48 max, audio true, 4 / 6 / 8 seconds. Quote them. Do not invent model URLs. The published Turbo link is Kling 3 Turbo Text to Video. The published Veo link is VEO 3.1.

Does a 3-second Kling V3 Standard take cost less than 51?

No. The row bills first 5 seconds at 51, then 11 a second after that. Durations include 3s, but the opening block does not shrink. Three, four, and five seconds are the same 51-credit floor. If the beat is a glance, generate the block you already pay for and cut in an editor. That is the same shape as other base_plus_per_second cards, with this row's own 51 and 11, not someone else's 70 and 14.