Riverflow 2.0 Fast lands in the catalog: ad-grade product shots in seconds

What's new

Sourceful has added Riverflow 2.0 Fast to the NetRoom catalog — a fast image model purpose-built for advertising, packaging and product scenes. It's the lightweight entry in the new Riverflow 2.0 lineup, aimed at commercial work where you need to keep the product recognizable while changing everything around it.

Why it exists

Plain text-to-image models render scenes beautifully but tend to fall apart on two things: keeping product geometry accurate and rendering readable text on labels. If you've ever dropped your bottle, coffee bag or box into Midjourney and asked for "the same, but on a beach," you know the result: the shape drifts, the logo reassembles itself, and tiny text turns into ornament.

Riverflow 2.0 Fast approaches the problem from the other side. You drop in one or more reference images, and the model preserves the packaging silhouette, colors, and label typography while placing it in the environment you describe. The result is a frame you don't have to fix in Photoshop afterward.

What it does

Riverflow 2.0 Fast covers four modes in a single model.

1. Scene from a description

Classic text-to-image. Describe the scene in words and the model builds it from scratch. Useful for concepts and moodboards before you have a finished product.

2. Scene from a photo

The main mode. Drop in up to four reference images and the model places your product in a new environment. You can mix a product reference and an environment reference: "bottle from here, background from there." If the model needs help with fine text or texture, there is a separate super-resolution reference slot — up to four more images used as a detail guide.

3. Edit an existing frame

Send in an image and ask the model to change a specific element: move the product to another table, change the lighting, repaint the background. Everything else stays put.

4. Detail-aware upscale

Raises resolution and sharpens small text, textures and edges. Useful when you need to push a draft into a print-ready or billboard-ready version.

Where it fits

• Marketplace listings. Shoot the product once, then put it into different environments — kitchen, outdoors, minimal, lifestyle.
• Ads for social and banners. Hero frames in 16:9, 7:3 or 9:16 — the model handles any aspect ratio.
• Brand catalog. Shoot one anchor frame, then build a consistent series in the same light and palette.
• Pre-production. Quick previews for client and studio sign-off before a real shoot.

Technical specs

• Resolutions: 1K and 2K across ten aspect ratios — from square 1:1 to ultra-wide 7:3 for billboards.
• References: up to 4 main + up to 4 super-resolution references for detail.
• Modes: text-to-image, image-to-image, edit, upscale.
• Output formats: JPG, PNG, WEBP.
• Speed: roughly half a minute per image — the model is tuned for throughput, not maximum quality.

What a real prompt looks like

To make this concrete — two examples of what you actually type in the description box.

A craft-soda ad shot:

"Premium beverage advertising shot: a craft orange soda bottle from the reference standing on a travertine slab, sliced blood oranges and curls of peel beside it. Condensation droplets on the glass, soft directional morning sunlight from the side, warm Mediterranean palette, clean editorial lighting. Keep the bottle shape and the readable label wordmark from the reference."

Attach a photo of your own bottle, and the model will mirror the silhouette and typography while assembling the rest of the scene from the description.

A cosmetics brand shot:

"Minimal skincare ad shot of a serum on pale grey marble. Soft diffused light, a gentle shadow from the bottle, a small eucalyptus sprig and a single oil drop nearby. Palette — ivory, stone grey, dusty green. Keep the bottle shape and silver cap from the reference, with a readable 'AURELIA' wordmark centered."

Notice that in both cases the prompt describes light, color, surroundings and styling — not the product itself. The model already knows the product from the photo; what you're describing is the world you want around it.

Tips for better results

• If you have a real photo of the product, always use it as a reference. Without it, the model has to "invent" the label and will often get it wrong.
• Describe environment, light and styling — not the product itself. Riverflow learns the product from the reference better than from any description.
• For wide hero frames, use 7:3 or 16:9 — they read more naturally than stretched squares.
• If you're heading for print, use the upscale mode after the scene is finalized. The model tops out at 2K, but detail can still be sharpened.

When not to pick Fast

Fast is the speed tier of Riverflow 2.0, not the heaviest. If your image needs to carry a lot of small readable text (a poster, magazine spread, infographic) or you're targeting very large-format print, wait for the Pro tier — we plan to add it to the catalog as well.

Try it on NetRoom

Riverflow 2.0 Fast is live in the NetRoom catalog. Open the model in your browser, drop in a product photo, describe the scene, and get a finished frame. No install, no proxy.