Seedream 5.5 is Dreamina’s model for creating and editing images. It can make a sticker with a translucent shadow, change the finish on a piece of furniture, lay out a poster around your copy, or restyle a portrait using a reference photo. What you bring to Dreamina depends on what you want to take away. A new illustration can start with a sentence. A revision to an approved design starts with the design itself.
Use the model selector in Dreamina’s image workspace to choose Seedream 5.5, then enter a description or add the image you want to edit.
What can you make with Seedream 5.5?
Match the model's capabilities to the asset you need. Each task starts with a different kind of instruction, even though the result is still an image.
Transparent assets suit stickers, decorative elements, and product visuals that need to sit against different backgrounds. Request that property explicitly rather than simply asking for an isolated object.
Targeted editing suits an image whose overall direction already works. You can ask for a change to one object, then make a further revision in another round. This gives an existing composition a new job: exploring a different product finish, adjusting an item of clothing, or removing an unwanted element.
Information-rich generation brings together content and visual organization. Posters, infographics, and presentation-style images need both the words and their hierarchy: which message is largest, what belongs together, and where supporting information goes. Supplying that structure gives the model a more definite design task than a string of style adjectives.
Portrait creation and editing cover photographic concepts as well as changes to makeup, hair, clothing, and accessories. Naturalness comes from visible qualities you can describe, such as skin texture, expression, light, and framing. A styling image can help you explore an appearance; it does not establish the physical fit of a garment or accessory.
A material change without a new scene brief
The official Seedream 5.5 example below shows an oversized sofa in a desert fashion scene. The product demonstration explores changing the sofa's color or material while asking to retain the rest of the scene.
The orange version has a glossy, inflated surface. The silver version adds mirror-like reflections of the terrace and desert. Both retain the recognizable arrangement of the sofa, two figures, circular stone structure, and dunes.
This makes the distinction between a color change and a material change concrete. A silver finish affects how the sofa reflects its surroundings, not just its hue. If you already have a useful composition, naming the object and its desired material lets you explore that finish within the existing visual direction.
For your own result, compare the features that matter outside the edited object before carrying the image into another round. A targeted edit can still affect details you intended to keep.
What do you actually get back?
The result is an image, including any text, objects, or interface elements depicted within it. That determines what you can do next.
For a transparent asset, RGBA adds an alpha channel to the red, green, and blue color channels. That channel controls which areas are opaque, fully transparent, or partially see-through. A PNG can contain transparency, but its filename alone does not prove that the background is transparent. A checkerboard painted into the picture is also different from an actual transparent area. Place the saved asset over a light and a dark background to see whether the background disappears and partially transparent details blend as intended.
For posters and presentation visuals, words drawn into the image are different from editable text boxes. Likewise, an image of a slide, website, or app interface is a visual design, not an editable slide deck, working website, or set of interface components. RGBA transparency does not turn those elements into separate layers.
If you need to update dates, prices, or translations repeatedly, keep the final text in an editable layout and use the generated imagery within it. If you only need a visual concept or a finished image, the generated page can serve that purpose without requiring a source document. Decide which deliverable you need before treating a good-looking preview as the completed job.
Start with a finish you want to change
Sign in to the Dreamina creator, open image creation, and select Seedream 5.5. Add a product photo you want to edit. For a chair, try: “Give the seat a brushed aluminum finish. Keep the shape of the seat, the legs, the camera angle, and the room.” Before generating, check the credit cost beside your chosen settings. If the workspace prompts you to change plans, open the offer to see which option includes that configuration.
Compare the new seat with the original at the edges and where the legs meet the floor. If the silhouette works but the reflections look too polished, use that version for a follow-up: “Soften the reflections on the seat to give the metal a satin finish.” If the chair’s shape changed, go back to the original photo and narrow the edit to the seat surface. For a new object, begin with its shape, material, and intended background in the prompt.
Questions to settle before your first image
How should prompts and reference images work together?
Use the prompt to assign the job; use references to show the visual information that the job depends on. State whether each reference supplies the subject, scene, color palette, style, or layout. A reference used for colors need not supply its people or setting too.
If your text asks for a different background from the one in a subject reference, make that separation explicit: retain the subject from that reference and describe the new setting. This resolves the intended roles before you ask the model to combine them.
Can you combine a person, product, and background from separate images?
Seedream 5.5's documented uses include combining multiple image references into a new composition. Identify which image supplies each element and explain their intended relationship: who holds or wears the product, where the subject belongs, and what should remain recognizable.
The result is a newly composed image. Judge scale, lighting, contact, and overlap together: a product that appears to float beside a hand has not fulfilled the intended relationship, even if both objects are present. If only one part needs correction, identify that relationship in the next edit rather than replacing the entire brief with a vague request to improve the image.
How do you request exact wording in a design?
Provide every word in quotation marks and specify its role, such as headline, subtitle, or footer. Include the positions and hierarchy that matter to the design. Then read the resulting text word by word, including dates and numbers. For a worked example, see how to correct a date and venue on an event poster.
For a correction, identify the text's location, quote the old wording and its replacement, and state which layout details should stay the same. Compare the corrected wording and surrounding layout before using the result. For repeated copy changes, use the editable-text approach described above.
Can you keep refining the same image?
Yes. Seedream 5.5 supports targeted changes across multiple editing rounds. Make each instruction self-contained: name the object, state the new change, and identify what should be preserved. “Change the sofa's material” is a more definite task than “make it like the earlier one.”
Keep the version you want to build from. If a later result changes a feature you need to retain, return to that image and give a narrower instruction. This keeps the next round anchored to the composition you actually chose.
For a reference-led composition, learn how to assign each image a clear role. For deliverable choices, use the guide to transparency and editable output.