Riverflow guide to editing product images and brand details

Riverflow is Sourceful's family of hosted image-generation and editing models for packaging, marketing, and other creative work. Sourceful introduced Riverflow 1 through its platform and Runware, and subsequently introduced Riverflow 2.0 with additional controls for typography and product detail. The published access routes are hosted services; the cited releases provide no open-weight download of the complete Riverflow system. Riverflow 1 announcement Riverflow 2.0 announcement

The practical question is whether an edit preserves the parts of a product that make it recognizable. This guide focuses on checking that requirement, with a documented Riverflow 2.0 Pro API example.

What are the key facts about Riverflow?

How does Riverflow handle product details and typography?

Riverflow 1's release describes an editing system developed around branded packaging and marketing work. Sourceful explains that its reasoning component can evaluate intermediate diffusion output and adjust an editing direction. That is the developer's description of the system, rather than evidence of a particular success rate on your own assets. Riverflow 1 technical overview

For Riverflow 2.0, Sourceful documents font control and reference-based super resolution. The latter uses additional reference artwork to repair details in a generated scene. These features are relevant when the correct label or product surface matters as much as the overall composition. Riverflow 2.0 features

For an evaluation, choose a product with recognizable details and keep a clean reference of its front, shape, and label. Write a short list of features that must survive the edit. Then request a change to the setting and compare those features directly with the source.

Separate aesthetic preference from product accuracy. A brighter scene can be more attractive while still showing an incorrect cap or altered lettering. Record both findings instead of collapsing them into a single impression of image quality.

What are the limits of Riverflow product editing?

Sourceful's first announcement explicitly says image-editing models remain short of the tight tolerances needed for physical print workflows. Treat an attractive packaging mockup as something to inspect, not automatic approval of production artwork. Riverflow 1 discussion

The complete system's availability is distinct from the components it uses. Sourceful says Riverflow 2.0 combines existing open and closed diffusion models with its own models and reasoning layer. That description does not constitute a downloadable Riverflow checkpoint or a license to reproduce the entire service. Architecture description

Version matters for integration. The documented Runware identifier in this guide is sourceful:riverflow-2.0@pro; controls and input fields should come from that model's API page. Do not mix a different Riverflow release's settings into a saved workflow without checking its documentation. Model request reference

Decide who approves the asset before generating a large collection. Give that reviewer the original product references, the requested change, and the candidate output. Ask them to identify concrete discrepancies rather than simply choosing the most visually appealing option.

How do you edit a product image with the Riverflow API?

Riverflow provides a hosted creative platform, and its models are also distributed through inference partners. For a repeatable developer example, use Runware's documented Riverflow 2.0 Pro route. Official platform information Provider examples

Create a Runware API key and set RUNWARE_API_KEY in your environment. Set PRODUCT_IMAGE_URL to a reference image you control that the provider can retrieve. This Python example uses the standard library to submit the documented request structure with a fresh task identifier:

import json
import os
import uuid
from urllib.request import Request, urlopen

job = {
    "taskType": "imageInference",
    "taskUUID": str(uuid.uuid4()),
    "model": "sourceful:riverflow-2.0@pro",
    "positivePrompt": "Place this product on a pale stone shelf. Preserve its shape, label, and colors. Use soft daylight.",
    "width": 1024,
    "height": 1024,
    "inputs": {"referenceImages": [os.environ["PRODUCT_IMAGE_URL"]]},
    "includeCost": True,
    "deliveryMethod": "sync",
    "outputType": "URL",
}
request = Request(
    "https://api.runware.ai/v1",
    data=json.dumps([job]).encode(),
    headers={
        "Authorization": "Bearer " + os.environ["RUNWARE_API_KEY"],
        "Content-Type": "application/json",
    },
)
with urlopen(request) as response:
    print(response.read().decode())

Runware documents reference images inside the inputs object, a unique task UUID, and optional cost reporting. Its examples show the same HTTP endpoint and authorization header. API schema HTTP examples

This request explicitly selects synchronous delivery and URL output. Inspect the returned imageURL and cost fields, and save accepted images with their task identifiers. If you change to asynchronous delivery, poll with getResponse instead of expecting the acknowledgment to contain an image. Delivery and response fields

Now compare the result in a fixed order:

1. Check the outline, proportions, and visible product parts.
2. Read the label against the supplied reference.
3. Check colors and the placement of distinctive markings.
4. Assess whether the new setting and lighting meet the brief.
5. Save only the versions the reviewer accepts.

If the setting works but the label fails, treat that as a specific unresolved requirement. The model documentation includes additional detail-reference inputs; consult those fields before testing a repair workflow. Preserve the accepted composition separately so the next experiment has a clear comparison point. Detail-reference fields

Keep the product source, scene instructions, and approved result together. For another campaign, reuse the source and review criteria while changing only the intended setting. This is a suggested working method, not a claim that the model guarantees identical output across requests.

How does Riverflow compare with Gemini 3 Pro Image?

The sibling-topic Gemini image overview explains the Google option. Compare both with the same product reference and editing instruction, paying attention to the details your reviewer considers mandatory.

If your alternative is operating an image workflow yourself, the relevant pillar is the ComfyUI complete guide. Make the deployment decision separately from the visual acceptance decision.

What else should you know about Riverflow?

Who makes Riverflow?

Sourceful develops Riverflow. Its official information page distinguishes the Riverflow marketing platform from Sourceful's packaging-design focus. Official company information

Is Riverflow an open-source image model?

Sourceful's cited releases describe hosted access without providing a download of the complete system's weights. Its architecture combines open and closed components with its own models. Riverflow 2.0 architecture

Can Riverflow preserve product labels?

Sourceful documents controls intended to improve typography and product detail. Use a reference and inspect the output against the actual label before approving the asset. Feature announcement

Which model does the API example use?

It uses Riverflow 2.0 Pro through Runware with the published identifier sourceful:riverflow-2.0@pro. Match the API reference to that identifier when adding inputs or changing delivery options. Provider documentation

Sources

• Sourceful's Riverflow 1 announcement
• Sourceful's Riverflow 2.0 announcement
• Official Riverflow company and platform information
• Runware Riverflow 2.0 Pro API reference
• Runware Riverflow 2.0 Pro examples
• Google Gemini 3 Pro Image model documentation

Related guides on PromptZone

• Best SDXL Models in 2026
• ComfyUI 2026: The Complete Guide
• AI Model Releases Timeline