Modes & Models

Choosing what to generate is a two-step decision. First you pick a mode — the capability, defined by what goes in and what comes out, like Image to Video. Then you pick the model that will do the work: each mode offers a curated set of the leading models for that job, each with its own strengths, options and fal.ai price.

Click the mode/model chip in the prompt bar (or the AI Modes and Models toolbar button) to open the Select Mode and Model dialog: capabilities on the left, model cards — with provider badge and description — on the right.

Image to Image

Transform a photo or viewport capture with a prompt: renders, try-ons, edits. This is the workhorse mode for jewelry — select an input image, describe the change, and keep what matters intact.

Model	Provider	Best for
Nano Banana Pro	Google	Final-quality results — hero shots, campaign imagery and detailed edits, up to 4K.
Nano Banana 2	Google	Fast, precise everyday edits that respect composition, lighting and style.
Reve Edit	Reve	Clean, natural edits with strong prompt adherence — a good second interpretation.
GPT Image 2	OpenAI	Fine-grained local changes: swap a stone, change a finish, keep the rest untouched.
Seedream 4.5	ByteDance	Photorealistic product edits with consistent subjects across variations.
FLUX.1 Kontext Pro	Black Forest Labs	Changing background, lighting or context while the piece stays identical.
Grok Imagine	xAI	Fast, creative takes when you want a different eye on the same edit.
BiRefNet v2	fal.ai	Background removal — clean cut-outs even on thin chains and prongs. No prompt needed.
Fibo	Bria	Licensed-data generations with fine control: steps, guidance scale, negative prompt.

Text to Image

Generate images from a text description: sketches, renders, concepts. No input image required — just describe the piece, the scene and the style.

Model	Provider	Best for
Nano Banana 2	Google	The default for jewelry concepts and renders from a plain description.
FLUX1.1 Pro Ultra	Black Forest Labs	Studio-photograph realism at up to 2K when the concept must look final.
FLUX LoRA	Black Forest Labs	Generating in a trained brand or product style (acts as standard FLUX dev without a LoRA).
FLUX.1 Schnell	Black Forest Labs	Rapid, very low-cost ideation — try many directions, refine the winner elsewhere.
Grok Imagine	xAI	Bold, creative interpretations when other models play it too safe.
Reve	Reve	A high-aesthetic second opinion on the same concept.

Image to Video

Create smooth, short videos from a single image — ideal for social media and e-commerce. Select a still (a render, a generation, a capture) and describe the motion.

Model	Provider	Best for
Kling 2.5 Turbo Pro	Kuaishou	The go-to for jewelry turntables and hero clips — top motion realism, 5 or 10 s.
Kling 3.0 Standard	Kuaishou	Current-generation Kling quality at a lower price than Pro.
Seedance 2.0	ByteDance	Fluid motion and strong subject consistency — compare with Kling on the same shot.
Veo 3.1	Google	Cinematic 8-second clips with natural physics and optional synchronized audio.
Wan 2.5	Alibaba	Volume work — animating many catalog shots at a strong price/quality ratio.

Text to Video

Generate short videos directly from a text description, with no source image at all.

Model	Provider	Best for
Kling 2.5 Turbo Pro	Kuaishou	Smooth 5 or 10 second clips with excellent motion, straight from a description.
Veo 3.1	Google	Cinematic scenes with natural physics and optional audio.
Wan 2.5	Alibaba	Cost-effective clips at 720p or 1080p.

Image to 3D

Produce an editable organic 3D model directly from a photo — or from a viewport capture of a shape you want reinterpreted. Results preview in the Studio’s 3D viewer and import into Rhino as meshes. See the workflow on Image to 3D by AI.

Model	Provider	Best for
Meshy 6 Preview	Meshy	Most control: topology, target polycount, symmetry, remeshing and texturing.
Hunyuan 3D 3.1 Pro	Tencent	Highest geometric detail on ornamental work, with PBR textures.
Hunyuan3D v3	Tencent	Quick, cheaper volume tests before rerunning the best candidates on Pro.
Rodin	Hyper3D	Clean, production-grade meshes that hold up to further editing in Rhino.
Seed3D	ByteDance	Fast one-shot reconstruction with strong fidelity to the photo.

Text to 3D

Generate an organic 3D model from a simple description, ready to refine or print. See the workflow on Text to 3D by AI.

Model	Provider	Best for
Meshy 6 Preview	Meshy	Textured meshes from a description, with art style, topology, polycount and symmetry control.

Video to Video

Turn a viewport or render video photoreal, or upscale a generated clip. This mode needs an item that already has a video — select one in the grid first.

Model	Provider	Best for
LTX Render to Real	Lightricks	Turning 3D render footage into photorealistic marketing video.
Topaz Video Upscale	Topaz Labs	Upscaling generated clips up to crisp 4K for e-commerce and social. No prompt needed.

---

Options

Every model exposes its own subset of options — open Model Options (the gear in the prompt bar, or the toolbar button) to see what the current model offers before you hit Send.

Option	What it does	Available values
Ratio	Aspect ratio of the generated image	Auto, 21:9, 16:9, 3:2, 4:3, 5:4, 1:1, 2:3, 3:4, 4:5, 9:16
Number of Images	How many candidates one generation produces	A number, 1 by default
Negative Prompt	What the model should avoid in the result	Free text
Duration	Clip length for Kling, Seedance and Wan videos	5 or 10 seconds
Veo Resolution	Output resolution of Veo 3.1 clips	720p, 1080p
Generate Audio	Whether Veo 3.1 adds synchronized audio	Yes, No
Wan Resolution	Output resolution of Wan 2.5 clips	480p, 720p, 1080p
CFG Scale	How strictly Kling follows the prompt	A number, 0.5 by default
Steps	Number of inference steps for Fibo	A number, 50 by default
Guidance Scale	Prompt-adherence strength for Fibo	A number, 5 by default
Mode	Meshy generation quality tier	Preview, Full
Art Style	Look of the Meshy text-to-3D result	Realistic, Sculpture
Topology	Mesh structure of the Meshy result	Triangle, Quad
Target Polycount	Mesh density Meshy aims for	A number, 300000 by default
Symmetry	Meshy symmetry handling	Auto, Off, On
Should Remesh	Whether Meshy rebuilds the mesh topology	Yes, No
Should Texture	Whether Meshy textures the image-to-3D result	Yes, No