Higgsfield charges 2 credits for a Nano Banana Pro image and 4 credits for the same image at 4K. Its own Soul 2.0 model costs 0.12 credits an image. Same subscription, same balance, a 16x spread between the cheapest and most expensive picture you can make.
Most people never notice, because the pricing page leads with video. But if images are half your output, the model you pick is the difference between 135 images a month and 2,250.
The Two Numbers That Matter
Higgsfield's plan cards quote image counts directly: Starter's 270 credits is listed as 135 Nano Banana Pro images, Plus's 1,200 as 600, Ultra's 3,000 as 1,500. That is the 2-credit rate applied straight down the line.
So the headline allowance is always Nano Banana Pro at 2K. Everything else is a multiplier on it:
| Model | Credits per image | Images from 270 credits | Images from 3,000 credits |
|---|---|---|---|
| Soul 2.0 | 0.12 | 2,250 | 25,000 |
| Soul / Reve / Product Placement | 0.25 | 1,080 | 12,000 |
| Wan 2.2 (image) | 0.5 | 540 | 6,000 |
| Nano Banana / Seedream / GPT Image | 1 | 270 | 3,000 |
| FLUX.2 Pro | 1 | 270 | 3,000 |
| Nano Banana Pro | 2 | 135 | 1,500 |
| Face Swap / Character Swap | ~2 | 135 | 1,500 |
| FLUX.2 Flex | 3 | 90 | 1,000 |
| Nano Banana Pro 4K | 4 | 67 | 750 |
| FLUX.2 Max | 4 | 67 | 750 |
The same 270-credit Starter pack buys 2,250 Soul 2.0 images or 67 Nano Banana Pro 4K images: a 33x swing in output for the same money. Model choice moves cost more than plan choice does.
Data table
| Item | Value | Share |
|---|---|---|
| Soul 2.0 | 0.12 | 0.7% |
| Soul / Reve | 0.25 | 1.4% |
| Wan 2.2 | 0.5 | 2.8% |
| Nano Banana | 1 | 5.6% |
| FLUX.2 Pro | 1 | 5.6% |
| Nano Banana Pro | 2 | 11.2% |
| Face / Char Swap | 2 | 11.2% |
| FLUX.2 Flex | 3 | 16.8% |
| Nano Banana 4K | 4 | 22.4% |
| FLUX.2 Max | 4 | 22.4% |
The 4K toggle doubles the price: 750 images instead of 1,500 on Ultra, for pixels most feeds will never display.
What an Image Costs in Dollars
Credits are an abstraction until you divide by the plan price.
| Plan | Cost per credit | Nano Banana Pro (2) | Nano Banana Pro 4K (4) | Soul 2.0 (0.12) |
|---|---|---|---|---|
| Starter $19 / 270 | $0.070 | $0.141 | $0.281 | $0.008 |
| Plus $47 / 1,200 | $0.039 | $0.078 | $0.157 | $0.005 |
| Ultra $99 / 3,000 | $0.033 | $0.066 | $0.132 | $0.004 |
A Soul 2.0 image on Ultra costs four tenths of a cent. A Nano Banana Pro 4K image costs 13 cents. Both are "an image" in the same interface, and the gap is 33x.
Where the Cheap Tier Holds Up
Soul 2.0 is Higgsfield's own model, and at 0.12 credits it is priced like a utility rather than a product. That is the tier for volume: character sheets, style exploration, background plates, thumbnails, anything you are going to generate forty of and keep three.
Nano Banana Pro is the tier for anything with text in it, product accuracy, or a reference image that has to survive intact. It is the model the plan cards advertise, and it is the one that holds composition when you feed it a photo.
The practical split: explore on Soul 2.0, commit on Nano Banana Pro. A 40-image exploration pass costs 4.8 credits on Soul and 80 on Nano Banana Pro. Same exploration, 16x the bill.
One caveat: Soul 2.0 is a Higgsfield model, not a frontier third-party one. If a client is paying for a specific look, the cheap tier is where you find out whether it can hit it.
The Video Comparison Nobody Makes
Here is the number that reframes the whole credit system. One second of Kling 3.0 at 720p costs about 1.4 credits. That is 11.7 Soul 2.0 images, or 0.7 of a Nano Banana Pro image.
Put differently: a single five-second Kling clip at 720p (7 credits) costs the same as 58 Soul 2.0 images. If you are budgeting a campaign and images are the deliverable, video is not the expensive part of Higgsfield. It is the part that eats the balance.
That is also why the image allowance on the plan cards looks generous. 1,500 Nano Banana Pro images on Ultra is 3,000 credits: the same balance that buys 75 four-second Veo 3.1 clips. Nobody generates 75 Veo clips a month, which is exactly why the image number reads as huge.