How to Use GPT Image 2 Image-to-Image: 1K/2K/4K Tiers, Credits and Prompts
A working guide to GPT Image 2 image-to-image on VdoBloom: 10 inputs, a 20,000-character prompt, eleven aspect ratios with auto, 1K/2K/4K at 3/5/8 credits, plus the two constraints that trip people up.
GPT Image 2 image-to-image on VdoBloom edits up to 10 attached photos from a single instruction of up to 20,000 characters, outputs at 1K, 2K or 4K across eleven aspect ratios including an auto mode that matches your source, and costs 3, 5 or 8 credits per image depending only on the resolution you pick. It is the successor to GPT Image 1.5 and it is both more capable and cheaper — a 1K edit costs 3 credits against 1.5’s flat 10.
What GPT Image 2 I2I is best at
Three things separate it from the rest of VdoBloom’s editing roster. The first is prompt headroom: 20,000 characters is not a typo, and it is roughly six times what most editors on the platform accept. You can paste an entire brand style guide, a shot list, and a do-not-change list into one instruction and the model will work through it. The second is the auto aspect ratio, which matches the output to your source image’s dimensions instead of cropping or padding it into a preset frame — the correct choice whenever the job is “change this one thing” rather than “recompose this picture.” The third is that resolution is a real dial, from a cheap 1K draft to a 4K final, on the same model with the same behaviour.
Practically, it is the strongest general-purpose choice on VdoBloom for e-commerce retouching, packaging and label work, marketing composites built from several reference photos, and any edit where legible text has to survive. It holds a 4.6 user rating, the highest of the GPT Image 2 generation on the platform.
Real specs and exact credit costs
The credit numbers below come from VdoBloom’s pricing table directly. Dollar figures use the Lite plan rate: $15 for 300 credits, so one credit is $0.05.
| Resolution tier | Credits per image | Cost at Lite rate | What you get |
|---|---|---|---|
| 1K (default) | 3 | $0.15 | Drafts, social crops, thumbnails, fast iteration |
| 2K | 5 | $0.25 | Web hero images, product listings, ad creative |
| 4K | 8 | $0.40 | Print, large displays, crop-and-zoom headroom |
| Spec | GPT Image 2 image-to-image |
|---|---|
| Reference images per job | Up to 10 |
| Prompt length limit | 20,000 characters |
| Aspect ratios | auto, 1:1, 5:4, 9:16, 21:9, 16:9, 4:3, 3:2, 4:5, 3:4, 2:3 (eleven) |
| Default aspect ratio | auto |
| Resolutions | 1K (default), 2K, 4K |
| Known constraints | auto aspect forces 1K output; 1:1 cannot be rendered at 4K |
| Content filter tier | Strict — OpenAI provider filters |
| User rating on VdoBloom | 4.6 |
Read those two constraints again, because they are the two things that surprise people. If you leave the aspect ratio on auto, you get a 1K result and you are billed 3 credits — the platform prices what actually rendered, so you are not overcharged, but you also do not get the 4K you thought you selected. And square 4K is simply not available; if you need a large square, render 4K at 5:4 or 4:3 and crop.
How to prompt it well
The model is an instruction follower first and an image generator second. Structure your prompt accordingly.
- Lead with the single change. One sentence stating the edit, before any styling detail. “Replace the background with a matte concrete wall.”
- Then write the preservation list. This is where the 20,000-character budget earns its keep. Spell out the product label text, the logo, the hands, the reflections, the shadow direction — anything unlisted is negotiable to the model.
- Label each attachment. With up to 10 inputs, say what each one is for: image 1 is the subject, image 2 is the colour reference, image 3 is the lighting reference.
- Draft at 1K, finish at 4K. Iterate the wording at 3 credits a go, then re-run the winning prompt at 8 credits. Ten drafts plus one final is 38 credits, $1.90.
- Choose auto only for targeted edits. Object removal, colour changes and retouching benefit from source-matching dimensions. Anything that needs a specific delivery frame should name that ratio explicitly so you are not silently pinned to 1K.
If the photo you are editing shows a real, identifiable person, you must have that person’s consent before you edit and publish their likeness.
A worked example
Goal: turn a plain catalogue shot of a running shoe into a 4K campaign image. Attach the catalogue shot, a photo of wet asphalt, and a swatch of the brand’s orange. Set aspect to 3:2 (not auto, because you want 4K), resolution to 4K, and write:
“Place the shoe from image 1 on the wet asphalt surface from image 2, shot from a low three-quarter angle. Preserve the shoe’s exact silhouette, stitching, sole tread pattern and all printed branding without alteration. Add a reflection of the shoe in the wet ground. The only accent colour in the scene should be the orange from image 3. Do not add people, text, or additional products.”
Four 1K drafts to settle the wording (12 credits) plus one 4K final (8 credits) comes to 20 credits, or $1.00. For comparison, the same workflow on GPT Image 1.5 I2I would cost 50 credits, because 1.5 bills a flat 10 credits per run with no cheap draft tier.
When to pick a different model
- You need more than 10 reference images. The GPT Image 2.5 Flare and Sunburst editors on VdoBloom accept 16 inputs at the same 3/5/8 credit pricing, and 2.5 also unlocks 27:16, 16:27, 9:8 and 8:9 frames (those four are 1K-only).
- Your content is swimwear, dance, fitness or fashion that mainstream tools refuse. Every GPT Image model sits in the strict tier because it inherits OpenAI’s filters. Use one of VdoBloom’s flexible-tier editors instead. Explicit and illegal content is blocked on every model.
- You are producing hundreds of near-identical variants. Even 3 credits adds up at volume; cheaper editors exist on the platform for work that does not need literal instruction-following.
Run it from the AI image editor, generate source images in text-to-image, borrow phrasing from the GPT Image 2 prompt library, and check plan maths on the pricing page. New accounts get 10 free credits with no card — three 1K edits before you spend anything.
Frequently asked questions
How many credits does a GPT Image 2 image-to-image edit cost?
3 credits at 1K, 5 at 2K and 8 at 4K, per image. At the Lite rate of $15 for 300 credits that is $0.15, $0.25 and $0.40 respectively. Text-to-image on the same model is priced identically.
Why did my 4K selection come out at 1K?
Because the aspect ratio was set to auto. Auto forces a 1K render, and VdoBloom bills the 3-credit 1K rate to match what actually rendered. Pick a named ratio such as 3:2 or 16:9 when you want 2K or 4K.
Can I render a square image at 4K?
No. The 1:1 ratio cannot be combined with 4K on this model. Render 4K at 5:4 or 4:3 and crop to square, or accept 2K for a native square.
Is GPT Image 2 better than GPT Image 1.5 for editing?
For nearly every job, yes. It carries a higher rating (4.6 against 4.4), takes a 20,000-character prompt instead of 3,000, offers eleven aspect ratios instead of three, adds 1K/2K/4K resolution control, and costs less at every tier. GPT Image 1.5 remains available if an existing workflow is tuned to it.
How long can my editing instruction be?
Up to 20,000 characters, which is the largest prompt budget of any image editor in this family on VdoBloom. Use it for explicit preservation lists rather than adjective stacking.
Do paid plans remove the watermark?
Yes. Paid plans download watermark-free with commercial rights, and one-time credit packs from $2.49 for 75 credits never expire if you prefer not to subscribe.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →