AI Easy Mode

AI Easy Mode is the path for non-developers. Upload a photo or write a short prompt; peg.fun does the rest. In under 5 minutes you sign one transaction and a complete tradeable on-chain collection is live, token, art, hook, bonding curve, and creator vault, all deployed atomically.

No rect data. No Solidity. No design software. Just a wallet and a photo.

What you need before you start

  • A Base mainnet wallet with ≥ 0.001 ETH for the AI service fee, plus some gas for the deploy transaction.
  • A photo (PNG, JPG, or WebP, ≤ 8 MB, ≤ 2048 × 2048) or a short text prompt describing the character. Photos give the most consistent results.
  • A name and a ticker symbol for your collection.

The flow, step by step

  1. Upload or describe. Drop a photo or type a prompt. You can see a small preview of the source as you type.
  2. Pay 0.001 ETH. A one-time off-chain service fee that unlocks a single AI session. The fee covers up to 96 generation calls and lasts for one hour.
  3. Sign the AI session. A single wallet signature authorizes peg.fun's backend to run the AI pipeline on your behalf. No gas, no ERC-20 approval.
  4. AI extracts layers. The pipeline uses SAM-2 segmentation to split your photo into layers, body, head, eyes, accessory, etc. It also proposes how many variants each layer should have (4-8 is typical).
  5. AI generates variants. For each layer, peg.fun generates pose- and theme-consistent variants. For a horn layer, you might get small / large / broken / golden across multiple poses.
  6. Edit and regenerate. Rename layers. Adjust variant counts. Pick which variants are common, rare, or legendary. If a slot doesn't look right, regenerate just that slot (each regen consumes one call from your 96-call budget).
  7. Preview. peg.fun renders a handful of sample composed objects using the same on-chain renderer that will run for real after deploy. What you see is what gets deployed, there is no AI between the preview and the chain.
  8. Deploy. One transaction, peg.fun's factory deploys the token contract, the v4 hook, the bonding curve, and the creator vault atomically. Within a few blocks your collection is live at peg.fun/t/<address>.

What's AI vs what's deterministic

The boundary matters because it's what makes the protocol reproducible.

AI (non-deterministic):

  • Photo stylization
  • Layer segmentation and naming
  • Variant generation (pose, theme, color suggestions)

Running these steps twice on the same input can produce different output. That's fine, the result is what you preview and approve. Once you click Deploy, the bytes are frozen.

Classical pipeline (deterministic):

  • Pixel quantization to peg.fun's fixed 36-color palette
  • Greedy rect-packing into compact on-chain bytes
  • Slot config serialization
  • On-chain SVG rendering

Same input bytes always produce a byte-identical output. This guarantee is what lets the indexer reproduce visuals off-chain without round-tripping to the chain, and what makes the rendering reproducible forever, even if peg.fun's frontend disappears.

Once the preview is displayed, the slot configuration blob is frozen. Clicking Deploy submits exactly those bytes. We don't re-run any AI between preview and deploy.

The 96-call budget

One 0.001 ETH session buys:

  • 1 segmentation call (one-shot, automatic).
  • 1 character decomposition call (one-shot, automatic).
  • 96 variant-generation calls (consumable, each "regenerate this slot" click costs one).
  • Unlimited deploy-pack attempts (free; no AI involved).

A typical creator uses 32-50 of the 96 calls (8 slots × 4-6 variants). The rest covers slot regeneration if a variant isn't right.

If you burn through the budget before deploying, you have two options:

  • Deploy with the current state (the variants you've already generated are still good).
  • Pay another 0.001 ETH for a fresh session.

What the 0.001 ETH fee covers

The fee goes to the peg.fun platform treasury. It is not part of the bonding curve liquidity, it pays for the AI provider costs (Replicate + OpenAI API calls) and adds spam resistance.

The fee is verified on-chain before any AI call runs. The platform checks:

  • Correct recipient (treasury wallet),
  • Sufficient amount (≥ 0.001 ETH),
  • Confirmed on Base mainnet,
  • Not already consumed by an earlier session.

What happens if AI fails

If a provider call fails, peg.fun retries automatically. If both Replicate and OpenAI are unreachable end-to-end, the session ends with an AI_PROVIDER_FAILED_NON_REFUNDABLE diagnostic.

The 0.001 ETH AI session fee is non-refundable. If providers fail end-to-end, the fee is forfeit. This is the trade-off peg.fun makes for the simplest possible operational footprint, no refund queue, no batched treasury payouts. See Risks for the long version.

Permissionless posture

Anyone with 0.001 ETH can launch. There is no per-wallet daily cap, fee economics handle spam. There is no admin pause, no graduation ceiling, and no time-window limits at the protocol layer. The on-chain factory.launch call is fully permissionless; the off-chain AI fee is the abuse-resistance mechanism.

See What is peg.fun? and How peg.fun works for the broader protocol context.

When AI Easy Mode is offline

If the platform's image-quality eval gate fails, the /create page shows a maintenance banner. This gate runs automatically, there is no admin toggle. The only thing that takes AI Easy Mode offline is a prompt update that degrades quality below the minimum threshold.

Related