The image model race just split into four lanes
→ find which model fits your work
Welcome back to Visually AI!
Today’s reading time is 5 minutes.
In today’s edition:
Stop asking which image model is “best” - the top 4
Seedream 5.0 Pro's layer trick
Image & Video Prompts
Stop Asking Which Image Model Is Best
Four serious image models are on the table right now: GPT Image 2, Nano Banana Pro, Seedream 5.0 Pro, and Ideogram 4. The most useful thing I can tell you is that “which is best” is the wrong question. They each have pros and cons:
GPT Image 2 is the thinker. It is natively inside ChatGPT but it is available via API on several platforms. It reasons about your prompt (and can search the web) before it draws, so dense text, infographics, and fact-heavy layouts come back right far more often. It tops both community leaderboards. However, it can be the most expensive of the four (depending on where you use it), and editing means re-prompting. Conversational edits work well, but there are no region tools, sketch inputs, or layers.
Nano Banana Pro by Google DeepMind is still the realism and editing workhorse utilizing up to 14 reference images, the best style matching of the group, and a free tier inside Gemini (output will have watermark). If you’re placing a product into a real scene or matching an existing look, I would choose NBP.
Seedream 5.0 Pro by Bytedance (brand new — more below) is the editor’s model. Sketch directly on an image to direct changes, get native text in 14 languages, fuse up to 10 reference images, and separate a finished image into layers.
Ideogram v4 went open-weight and it’s available on the Ideogram web app. It’s the design specialist for posters, typography and brand assets. You can download it, run it locally, and even train it on your own style.
You can use these models on their parent app or platform (ChatGPT, Gemini, Dreamina/CapCut, or Ideogram) or on aggregator platforms like Fal, Krea, or Magnific. Honestly, I have found all 4 to be useful for various reasons, including using multiple reference images, creating character or prop sheets, or cinematic stills. Use what you already have available and don’t hesitate to refine prompts to explain exactly what type of output you want.
Here’s the whole picture in one table:
Seedream 5.0 Pro: Layers, Not Just Pictures
ByteDance released Seedream 5.0 Pro on July 8, and layer separation was a big introduction.
Every AI image you’ve ever generated is normally one flat picture. So if you loved the layout but wanted to nudge the title - reprompt, regenerate and hope. Seedream 5.0 Pro can split a finished image into separate elements: background, characters, text, and objects each on its own transparent layer (anywhere from 2 to 20). You’re getting a working design file you can open in Figma or Photoshop.
Pair that with sketch editing (draw rough marks on an image and it reads them as instructions) and you regenerate far less and finish far more.
However, Seedream 5.0 Pro is still capped at 2K resolution for now, and in side-by-side testing Nano Banana Pro still edges it on pure realism.
🖼️ Image Prompts
Prompt: A single illustrated page from a vintage field guide titled "URBAN BEEKEEPING — A BEGINNER'S FIELD GUIDE", hand-drawn in watercolour and ink on cream paper. Layout: a labelled cutaway diagram of a rooftop hive in the centre with eight small caption labels (brood box, queen excluder, super, frames, entrance, landing board, roof, stand), a honey-harvest calendar strip along the bottom listing May through September, and three numbered safety notes in the right margin in small handwritten script. Muted palette of ochre, honey gold, and faded teal. Soft even lighting, visible paper grain, every word sharply legible.
Prompt: A two-color concert poster for a fictional brass band, “THE PORTLAND LOW NOTES”, in bold art-deco typography. Hierarchy: band name as the dominant headline in the top third, tagline “An Evening of Brass & Velvet” in smaller caps beneath it, date and venue block at the bottom reading “SATURDAY 22 AUGUST — THE ALBANY THEATRE”, a geometric sunburst ornament behind the headline. Burnt orange and deep navy on warm cream stock, subtle ink-print texture, flat graphic style, all text crisp and correctly spelled.
Prompt: A high-resolution, wide shot of a museum-quality wall exhibit featuring a perfectly organized, educational solar system poster. The infographic displays the sun on the far left, followed by the eight distinct, non-duplicated planets in order: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, and Neptune. Each planet is rendered with unique textures, rings, and moons. Sharp, legible sans-serif labels are placed next to each planet with specific scientific facts and distance-from-sun data. The design is clean, academic, and minimalist on a matte black background, with perfect typographic clarity. Photorealistic, 8k.
🎥 Video Prompts
Text to Video Prompt: A young bicycle courier races through a rain-soaked vertical city at midnight, balancing a stack of steaming noodle containers inside a transparent delivery case. Track beside her as she pedals through dense pedestrian traffic, narrowly passes a street-cleaning robot, then descends a glowing spiral ramp into a crowded lower market. The camera transitions smoothly from a lateral tracking shot into a low front-facing angle without changing her direction or position. Neon signs reflect across wet pavement, steam rises from food stalls, umbrellas move naturally through the crowd, and passing vehicles create realistic wind displacement. Energetic cyberpunk street realism with rich color separation and crisp environmental detail. Sound: bicycle gears, rain, market chatter, distant transit announcements, electronic percussion. No readable text.
Seedance 2.0 Fast on Magnific
Image to video prompt: Handheld 50mm, warm neon + red-lantern bokeh, glass door beside him. Lock face, cap, beard, glasses. 0-2s: he holds his level gaze at camera, slow blink. 2-4s: faint exhale, slight head turn, his expression eases a touch. 4-5s: neon bokeh shifts, blurred passerby crosses far behind. Micro head-sway only. Keep beard + frames sharp. Movie-level stable features, detailed grey beard.
Kling 3.0:
🎨 Explore my Portfolio
I’ve had some amazing opportunities to work on a variety of projects recently, and I don’t share everything in this newsletter or online. And you know I love testing tools to share.
Take a look at my portfolio to get a quick glimpse of my work:
🚀 My Recent Top AI Tools Picks
Seedream 5.0 Pro — layer separation and sketch editing
Ideogram v4 — open-weight typography and design
GPT Image 2 — text-heavy and research-grounded images
Midjourney v8.2 - you can generate images with Midjourney’s next model by typing —preview at the end of your prompt
Generated with Midjourney v8.2 --preview
Thank you for reading.
I hope you have a creative week!
Heather Cooper