Krea 3 Goes Live on fal With 8-Image Editing at $0.025 Per Megapixel
A.I. / news
Krea 3 Goes Live on fal With 8-Image Editing at $0.025 Per Megapixel
The editing-focused image model was previewed on Aug. 28; no model card, licence or independent benchmark has surfaced.
Krea's Krea 3 image model went live through fal's API on Oct. 15, with an edit endpoint that takes one to eight reference images and a price of $0.025 per megapixel. Krea previewed the model on Aug. 28 and said then that it was not generally available.
The fal page, Krea 3 on fal, is the only primary description of the model this reporting could read. Krea's own site was not retrievable. It lists no parameter count, licence, model card or benchmark, and nothing in either source says the weights are downloadable.
What the API offers
fal describes Krea 3 as a model for reference-driven generation and editing, with three endpoints: edit, text-to-image and realtime. The edit endpoint combines several reference images into one while keeping details such as a face, a product or a fabric. It is also pitched at keeping a character or outfit consistent across a sequence, swapping materials or paint colours, and turning sketches into photorealistic images.
There are two variants. "Berry" is the default and is quality-focused. "Jam" is speed-focused. Reasoning levels are adjustable.
| Limit | Value on fal |
|---|---|
| Reference images (edit) | 1 to 8 |
| Maximum output | 4,220,160 pixels, such as 2048×2048 |
| Side length | 256 to 3,200 pixels, rounded up to multiples of 16 |
| Realtime frames | 768×768 JPEG |
What it costs
Text-to-image and edit both cost $0.025 per megapixel, counting the output and every reference image. A 1024×1024 image is $0.025. High reasoning on the edit endpoint adds 5 megapixels to the billed total. Billing is pay-as-you-go, with no subscription or minimum.
By that rule, an edit with two 1-megapixel references and a 1-megapixel output bills 3 megapixels, or $0.075. Switching on high reasoning bills 8, or $0.20. Those two figures are arithmetic from fal's published rate, not prices fal quotes.
- 1MP output, no reference0.03 USD
- 1MP output, two 1MP references0.07 USD
- Same, high reasoning0.2 USD
Source: Calculated from the fal.ai Krea 3 pricing page, accessed 2026-10-16
The August preview and the agents
AlphaSignal, which covered the Aug. 28 preview, said Krea called Krea 3 an in-house foundation model focused on editing, meant to carry on the look of Krea 2. The outlet wrote that Krea Edit features such as relighting may be folded into the base model.
Krea showed a second product the same day, Krea Agents, a platform where agents drive the canvas instead of a node graph. AlphaSignal said Krea spent three months building agents for creative work rather than developer tasks. It said both products would integrate through MCP or API from day one, and that Krea's MCP server already works with Claude, Codex, Cursor and OpenClaw, with access to more than 60 image and video models.
What is unverified
No benchmark, human preference study or sample set from an independent party is cited by either source. fal's feature list is a vendor description. Neither source says what Krea 3 was trained on.
Language-model releases such as Meta's Muse Spark Open and Liquid AI's LFM2.5-24B-A2B come with downloadable weights. The sources describe Krea 3 only as a hosted model.
Krea Agents has no availability date in the material reviewed.
Sources
More in A.I.
- 01Qwen3-4B-Deslop Cuts Its Slop Score From 121 to 12.5 on a MacBook Air, and Two of 40 Answers Came Out DamagedThe Show HN fine-tune by Erik Markus Kannike uses a rank-16 LoRA and a 175-byte neuron edit; its blog gives no licence, its Hugging Face card says Apache-2.0.
- 02HybridAgent Paper Claims 10.4% Gain and 2.7x to 3.7x Lower Cost, but Its Own Text Gives 10.4% Two Different MeaningsThe Stanford, UIUC and Together AI preprint ships MIT-licensed 8B and 14B agents, and no independent run of its numbers has been reported.
- 03Gemini 4 Argon Reaches the API at $2 and $10 Per Million Tokens, With a No. 1 Ranking and No ScoreGoogle's October 13 post says Argon tops the Vals Index but publishes no figure for it, and the introductory price doubles on November 30.
- 04DeepSeek Swapped V4.1-Flash-1017 Behind Its API Name, and the New Card Lists NL2Repo at 65.4 Where the Old One Said 64.0The MIT-licensed update calls itself minor and is now what deepseek-flash serves, with no active-parameter count and no release date on the card.