GPT Special: AI Image Generator & Editor Online
Bottom line
GPT Image 2 at 2K plus an automatic WAN 2.7 pass — 20 credits, 12 refunded if no edit is needed. Generate from text or edit an existing image with GPT Special in one place — VdoBloom shows the credit cost for either mode before you confirm.
GPT Special ("special/gpt" in VdoBloom's model list) is a two-stage pipeline, not a single model. Your prompt goes through an LLM router first, which splits it into a base description and, if there's one, a refinement instruction. Stage one generates the base image on GPT Image 2 at fixed 2K resolution. Stage two, when there's something to refine, runs that result through WAN 2.7 Image, also at 2K, for the precision edit — a color grade, an added element, or content the router decided should go to the more permissive editing stage rather than the stricter base model.
The billing follows the pipeline honestly: VdoBloom charges 20 credits upfront, and if the router determines there's nothing for stage two to do, the 12-credit edit component is refunded automatically — so a straightforward generation with no refinement needed nets out to 8 credits, not 20.
It functions as both generator and editor — upload an image and describe a change, or start from text alone. Eight aspect ratios span the practical range: 1:1, the 2:3/3:2 and 3:4/4:3 pairs, 9:16, 16:9, and 21:9. Despite running on GPT Image 2 underneath, it's tagged HOT rather than STRICT, since sensitive content gets routed to the more permissive WAN stage. Rated 4.6 by VdoBloom users, released 2026-04-21.
How the two-stage pipeline works
An LLM splits your prompt into a base description and an optional refinement instruction. Stage one generates the base on GPT Image 2 at fixed 2K. Stage two, if triggered, runs WAN 2.7 Image at 2K on top of that result. The split exists partly for content handling: GPT Image 2's base pass stays SFW, while anything sensitive routes to WAN 2.7's more permissive stage instead of failing outright at the base model.
What it costs on VdoBloom
20 credits charged upfront, 12 refunded automatically if the refinement stage is skipped — net 8 credits for a generation-only result, or 20 if the refinement runs. A full pipeline failure refunds the entire charge.
GPT Special vs GPT Image 2
GPT Image 2 alone is resolution-tiered (3/5/8 credits at 1K/2K/4K) with no refinement stage and a STRICT content tier. GPT Special fixes the base at 2K, adds an automatic WAN 2.7 refinement pass, and relaxes to HOT — at a higher nominal price (20 credits, net 8–20) that reflects the extra pipeline stage rather than a pricier base model.
What creators use GPT Special for
- Product hero images where on-image headline text has to be spelled correctly the first time.
- Uploading a rough draft image and requesting targeted edits like swapping a background or changing a garment color.
- Ultrawide 21:9 concept frames for pitch decks alongside 9:16 verticals for the same campaign.
How to use GPT Special on VdoBloom
- 1Open the Text to Image tool and select GPT Special.
- 2Write your prompt (or upload an image to edit).
- 3Set aspect ratio, review the credit cost, and generate.
Frequently asked questions
How much does GPT Special actually cost?
20 credits are charged upfront. If your prompt has nothing for the refinement stage to do, 12 credits are refunded automatically, netting 8 credits for a base-only generation. If the refinement stage runs, the full 20 stands.
What does "auto-enhanced" mean on GPT Special?
It's a two-stage pipeline: GPT Image 2 generates a base image at 2K, and — when your prompt calls for it — WAN 2.7 Image applies a precision refinement pass, also at 2K. An LLM router decides how to split your prompt between the two stages.
Is GPT Special free to try?
Your 10 free signup credits don't cover the full 20-credit charge on their own, but if the edit stage is skipped and 12 credits refund back, you're left with a net 8-credit charge — meaning 10 free credits can cover a base-only generation with 2 left over.
Can GPT Special edit images I upload, or only generate new ones?
Both. It's listed as a generator and editor, so you can start from text alone or upload an image and describe the modification you want. The same two-stage pipeline and pricing applies either way.
Which aspect ratios does GPT Special support?
Eight: 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, and 21:9 — the intersection of what GPT Image 2 and WAN 2.7 Image both support, since a job can pass through either or both stages.
How strict is GPT Special's content filtering?
GPT Special carries VdoBloom's HOT label — a notable departure within its own family, since the standalone GPT Image 2 editor sits in the STRICT tier. The router sends sensitive or suggestive material to the more permissive WAN 2.7 refinement stage rather than through GPT Image 2's stricter filters directly, though explicit and illegal material remains blocked platform-wide.
Why is GPT Special less strict than plain GPT Image 2?
Because of how the pipeline splits your prompt: GPT Image 2 only ever sees a cleaned-up, SFW base description, while anything more sensitive is handled by WAN 2.7 Image in stage two, which carries VdoBloom's more permissive HOT label.
GPT Special vs plain GPT Image 2 — why would I pick Special?
GPT Special adds an automatic refinement pass and more flexible content handling for 20 credits (netting 8 if unused), versus GPT Image 2's resolution-tiered 3–8 credits with no refinement stage. If you want one submission that both generates and polishes, Special saves a manual second edit step; if you just need a clean generation, GPT Image 2 alone is cheaper.
Does GPT Special always run the WAN 2.7 refinement stage?
No — only when the router finds something to refine in your prompt. If it doesn't, stage two is skipped and its 12-credit component is refunded, so you're not paying for a refinement pass that never ran.