Generate any poster, KV, or cover image with GPT Image 2.
Pick PSD as the output format and we'll auto-assemble a layered, editable Photoshop file — subject cutout, OCR text layers, and an inpainted no-text base.
Experimental: text removal and OCR accuracy vary; your PNG is always saved as a fallback.
Free credits for new users — no credit card required
A flat PNG can't be reverse-engineered into the original Photoshop source — fonts, vector paths, and asset files just don't exist in a flattened image. What this page does instead is build a *practical* layered PSD from the moment you generate: the original is preserved as a hidden reference layer, the same pixels are duplicated as a clean editable base, and text/FX groups are stubbed out so you can drop in copy and effects in Photoshop without untangling a single merged layer.
Every PSD ships with the same predictable structure: hidden REF, visible ART base, hidden CUT slots, visible FX group, editable TXT group, hidden NOTES. Open it in Photoshop and you know exactly where to edit.
The TXT group ships with a placeholder Type Layer so you can edit text in Photoshop directly — no rasterised baked-in copy. Replace the placeholder, change fonts, restyle freely.
We don't pretend a flat PNG hides original layers it never had. The hidden NOTES layer documents what's reconstructed and what's approximated, so handoff to a designer is transparent.
PSD assembly runs in your browser via ag-psd — no upload, no queue, no extra latency. Click PSD, file lands in your downloads folder.
Built for one job: generate, then hand the file off to Photoshop with structure intact.
Same engine the rest of the site uses — high-fidelity poster, KV, and cover output, no quality compromise to enable PSD export.
Group naming follows a fixed taxonomy (01_REF / 02_ART / 03_CUT / 04_FX / 05_TXT / 06_NOTES) so designers and automation scripts know exactly what to expect.
The original generated image is preserved as a hidden REF layer, and the prompt that produced it is written into the NOTES layer — so anyone opening the PSD later can trace its origin.
Empty CUT and FX groups give you a place to add subject masks, glow, smoke, or vignettes in Photoshop without restructuring the file.
Trial users see the standard watermark unlock dialog instead of a free PSD download — same paywall behaviour as the PNG download, no separate gate to maintain.
Common questions about the layered PSD export. More questions? Email support@imagesv2.ai.
The layer extraction pipeline runs three AI models in sequence — subject segmentation, OCR text detection, and text-region inpainting. Each step has its own failure modes: rembg may not perfectly separate complex subjects, OCR can miss highly stylised fonts (especially decorative Chinese typography), and the inpaint model occasionally leaves visible texture seams where text used to be. We mark this Beta until we can guarantee consistent quality across all poster styles. Your generated PNG is always preserved as a fallback — at worst the PSD layers are just less useful than expected. If a flat PNG poster is all you need, our Poster Generator produces one directly without the layer pipeline.
Yes — it's a standard binary PSD that Photoshop, Photopea, Affinity Photo, and any other PSD-aware tool can open. It's assembled in your browser via the open-source ag-psd library, not a fake .psd-renamed PNG.
The TXT group ships with placeholder Type Layers. Most PSD readers (including modern Photoshop) recognize them as text and let you edit content/font/size. A handful of older Photoshop versions may rasterise them on first open — when that happens, retype on a new layer; the structure still holds.
A single flattened image doesn't carry that information — the model only outputs a final composite. Auto-segmenting on the client is technically possible but slow and unreliable. Instead the PSD ships with empty CUT_* slots so you (or a Photoshop action) can drop a subject mask in without restructuring the file. We're evaluating an optional server-side cutout step (rembg / SAM via Replicate) for a future paid tier.
So you always have an undo path. If you over-edit the visible artwork, toggle the REF group on and you've got the pristine source pixels to clone from. It also makes the PSD self-documenting — anyone opening the file can see exactly what the AI produced before any edits.
If you generate with the Transparent background option enabled, the alpha channel survives into the ART layer. Otherwise the ART layer is a solid composite — you can mask it in Photoshop, or regenerate with transparency on.
Roughly 2-3x the size of the source PNG, because the original pixels are stored twice (once hidden as REF, once visible as ART) plus a small amount of overhead for the layer structure. A 1024×1024 PNG that's 1.5 MB typically yields a 4-6 MB PSD.
Same paywall as the PNG download. Trial users see the watermark unlock dialog when they click PSD; paid users get the file directly. No separate gate.
Two reasons. First, the assembly is genuinely fast in the browser — under a second for a 1024×1024 image. Second, it keeps cost flat: there's no per-export server fee, so we can offer PSD export as a free upgrade for any paid plan instead of metering it.
Free credits for new users. Generate any poster, click PSD, drop the file straight into Photoshop.





Quality Comparison (gpt-image-2)
| Quality | Speed | Image Detail | Credits/Image | Best For |
|---|---|---|---|---|
| Low | Fastest (3-8s) | Good composition, less detail | 10 | Quick iterations, bulk generation, social media |
| Medium | Moderate (10-20s) | Rich details, good textures | 40 | Marketing images, presentations |
| High | Slower (20-40s) | Highest fidelity, finest details | 110 | Print, posters, premium assets |
| Auto | Model decides | Auto-selected by model | 40 | When unsure |
Model Comparison
| Model | Highlights | Low/Image | High/Image |
|---|---|---|---|
| gpt-image-2 | Latest, best results | 10 | 110 |
Cost-Saving Tips
Enter a text prompt describing the image you want. Adjust parameters like size (Square, Landscape, Portrait), quality (Low/Medium/High), output format (PNG/JPEG/WebP), and background (Opaque or Transparent). Click "Generate" and GPT Image 2 will create a brand-new image from your description.
Tips for Better GPT Image 2 Results
Upload a source image, write a prompt describing the changes you want, and GPT Image 2 will modify the image accordingly. Without a mask, GPT Image 2 decides which areas to change. With a mask, you can precisely control which regions are modified.
You can use Edit mode for GPT Image 2 image-to-image generation without a mask. Simply upload a reference image and describe the transformation you want in the prompt — for example, "Convert this photo to a watercolor painting" or "Reimagine this scene in a cyberpunk style". GPT Image 2 will use your image as a reference to generate a new version.
Example GPT Image 2 Prompts for Image-to-Image