On 8frame, one press of generate on an image node returns one image on most models, and you pay for that one image. Three models let you ask for more and charge for each: Nano Banana 2 and GPT Image take 1 to 4, Qwen Image 3 takes 1 to 6, and all three start at 1. Midjourney works the other way: it always returns four images and charges once for the run, 32 credits on V8 (8 an image) and 28 on V7. Everything else, from Nano Banana Pro to Reve, returns a single image per run and has no count setting; the one exception is Seedream's layer mode, covered below. All of this is read from 8frame's tool code on 2026-10-01.
TL;DR
- Most models return one image per run and have no count setting: Nano Banana Pro and Classic, Seedream (apart from its Layerize mode), Imagen, FLUX.2, Grok Image and Reve
- Three models bill per image, at a count you choose: Nano Banana 2 (1 to 4, 11 each at 1K), GPT Image (1 to 4, 29 each at the default high quality, square) and Qwen Image 3 (1 to 6, 6 each at 1K). Each starts at 1
- Midjourney bills per run of four: 32 on V8, the default. You cannot ask for fewer
- From three images up, a Midjourney run is cheaper than Nano Banana 2 at 1K: three Nano Banana 2 images are 33
- Qwen Image 3 at 1K undercuts both Midjourney runs per image at any count: four for 24, six for 36
- Through the MCP connector, Midjourney is the only image tool that returns more than one image per call
Images per run, every image model
Credits. A credit is about $0.01 at pack rate. "Full run at the default setting" is the node untouched: its default model, resolution, quality and count.
| Model | Images per run (default / max) | Can you change it | Billed per image or per run | Full run at the default setting |
|---|---|---|---|---|
| Midjourney V8 (the Midjourney default) | 4 / 4 | No | Per run of four | 32 (8 an image) |
| Midjourney V7 (experimental) | 4 / 4 | No | Per run of four | 28 (7 an image); experimental, check it runs before relying on it |
| Nano Banana 2 (the Nano Banana default, 1K) | 1 / 4 | Yes, 1 to 4 | Per image | 11; four images 44 |
| Nano Banana Pro | 1 / 1 | No | One image, one charge | 21 |
| Nano Banana Classic ("Standard" in the picker) | 1 / 1 | No | One image, one charge | 6 |
| GPT Image 2 (the GPT Image default, high quality, 1:1) | 1 / 4 | Yes, 1 to 4 | Per image | 29; four images 116 |
| GPT Image 1.5 (high quality, square) | 1 / 4 | Yes, 1 to 4 | Per image | 18; four images 72 |
| Qwen Image 3 (1K) | 1 / 6 | Yes, 1 to 6 | Per image | 6; six images 36 |
| Seedream 5 Pro (the Seedream default, 1K) | 1 / 1 | No | One image, one charge | 7 |
| Seedream 5 Pro Layerize | A base image plus up to 16 layers; the model decides | No | Per run, charged as 17 layers | 78 at 1K |
| Imagen 4 (the Imagen default) | 1 / 1 | No | One image, one charge | 8 |
| FLUX.2 [pro] (the Flux default) | 1 / 1 | No | One image, one charge | 5 |
| Grok Image 2.0 (1K, no references) | 1 / 1 | No | One image, one charge | 9 |
| Reve (text to image) | 1 / 1 | No | One image, one charge | 34 |
Seedream 5 Lite and Seedream 4, Imagen 4 Fast and Ultra, and FLUX 1.1 Pro also return one image per run. Their prices, and every resolution and quality step, are in AI image generator price per image; what each node starts on is in AI image default settings by model.
How the count is billed
On the three count models, the price is the per-image price times the count, and every other priced setting multiplies with it. Nano Banana 2 at 2K is 17 an image, so four are 68. GPT Image 2 at low quality is 2 an image, so four are 8. Qwen Image 3 at 2K is 11 an image, so six are 66. The code clamps the count it bills: at most 4 on Nano Banana 2 and GPT Image, at most 6 on Qwen Image 3.
On Nano Banana, the count belongs to Nano Banana 2 only. The node sends a count only when Nano Banana 2 is the selected model. Nano Banana Pro and Classic return one image per run and bill one image (Pro is 21, or 42 at 4K).
Midjourney is the one per-run model. The node always asks for four images, and the version sets the price, not the count: 32 on V8, the default (V7, which the picker marks experimental, is listed at 28). There is no setting for fewer, so a single keeper still costs the whole run.
Seedream 5 Pro Layerize bills on what the model might return. It takes one image and returns a base image plus up to 16 separated layers, and the model, not the request, decides how many. 8frame charges for the full 17 up front: 78 credits at 1K, 155 at 1.5K or 2K. Refunds on 8frame are for a failed or cancelled run and return the whole charge; there is no partial refund when fewer layers come back.
Per image vs per run: which is cheaper for exploring
Per-run pricing is a bulk price: it pays off only when you look at all four images. Per-image pricing lets you buy exactly the number you want. At the default settings from the table:
- One image: FLUX.2 [pro] 5, Qwen Image 3 6, Nano Banana 2 11. Midjourney still charges the full run for the one you keep, 32 on V8.
- Two images: Nano Banana 2 at 1K is 22, under either Midjourney run.
- Three images: Nano Banana 2 at 1K is 33, over one Midjourney V8 run (32). From three candidates up, Midjourney's run costs less per image than Nano Banana 2 at 1K: 8 an image against 11.
- Four images: Qwen Image 3 at 1K 24 (6 each), Midjourney V8 32 (8 each), Nano Banana 2 at 1K 44 (11 each), GPT Image 2 at its default high quality, square, 116 (29 each).
- Six images: Qwen Image 3 at 1K, 36 in one run. Two Midjourney V8 runs give eight images for 64.
The one-image models repeat by hand, at the same price per image: four runs of FLUX.2 [pro] are 20 credits, four of Seedream 5 Pro are 28. The cheapest set of four from a single run is GPT Image at low quality with the count at 4, 8 credits; low quality is a drafting setting.
None of this says which images you will want to keep; it is only what each route costs. A sensible split: when you are exploring compositions and want many candidates from one prompt, use a per-run model or a high count on a cheap per-image model; when you already know the shot, a single image from a per-image model costs less, and the keeper can be rerun on the model the final needs.
Through the MCP connector
8frame's MCP connector, the route for Claude and other MCP clients, has no count field on any of its image tools, so every call returns one image. The exception is Midjourney: four images per call for 28 credits. The connector passes no Midjourney version, and 28 is the V7 price, where the canvas default, V8, is 32.
What differs from the canvas:
- Qwen Image 3: one image per call, 6 at 1K or 11 at 2K. The 1 to 6 count exists on the canvas only. Four images take four calls and cost 24, the same total as one four-image run on the canvas, because the price is per image either way.
- Nano Banana: the connector has no model field and charges 6 credits, the Nano Banana Classic price. Nano Banana 2, with its 1 to 4 count, and Nano Banana Pro are canvas-only.
- GPT Image: not one of the connector's image tools, so its count setting exists only on the canvas.
- FLUX.2 [pro], Seedream, Imagen, Grok Image, Reve: one image per call, as on the canvas. Seedream through the connector is 5 Pro at 1K, 7 credits, with no resolution field.
Setup, and what every connector tool costs, is in how to generate AI video from Claude.
FAQ
How many images does one generation make on 8frame? One on most image models. Nano Banana 2 and GPT Image make 1 to 4 and Qwen Image 3 makes 1 to 6, at a count you set, starting from 1. Midjourney always makes four.
Does asking for four images cost four times as much? On Nano Banana 2, GPT Image and Qwen Image 3, yes: the charge is the per-image price times the count. On Midjourney, no: the price covers the run of four.
Why did Midjourney cost 32 credits when I wanted one image? Midjourney is billed per run, and every run is four images: 32 on V8, the default. There is no one-image option.
Can Nano Banana Pro make four images at once? No. The count setting applies to Nano Banana 2 only; Nano Banana Pro returns one image per run, 21 credits (42 at 4K). Run it again for another.
What is the cheapest way to get four candidates from one prompt? In one run: GPT Image at low quality with the count at 4, 8 credits. Among the defaults in the table: four separate FLUX.2 [pro] runs, 20, or one four-image Qwen Image 3 run at 1K, 24; one Midjourney V8 run is 32.
If a multi-image run fails, do I pay for it? No. Failed generations are refunded automatically, and the refund is the whole charge for the run. See do failed AI generations cost credits.
Can I choose the number of images through the MCP connector? No. The connector's image tools have no count field: one image per call, except Midjourney at four for 28.
Sources
- Images per run, count controls and defaults (the number-of-outputs control on Nano Banana, GPT Image and Qwen Image 3, the count sent only for Nano Banana 2, GPT Image clamped to 1 to 4, Qwen Image 3 to 1 to 6, Midjourney fixed at four, one output on every other image tool): 8frame front-end tool registry, read 2026-10-01
- Whether the count multiplies the price (Nano Banana 2 per image up to 4, GPT Image per image up to 4, Qwen Image 3 per image up to 6, Midjourney a flat price per run, FLUX.2 [pro] and Grok Image billed and delivered as one image, Seedream 5 Pro Layerize charged as 17 layers, refunds only of the full charge): 8frame back-end tool registry and its credit cost functions, read 2026-10-01
- Connector behaviour (no count field, Qwen Image 3 one image per call, Midjourney four per call at 28, Nano Banana at 6 with no model field, no GPT Image tool): 8frame MCP connector tool definitions, read 2026-10-01
- Credit value and plan price: 8frame billing catalog
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.