Imagen 4 or Flux for photorealism?

Imagen 4 Ultra wins on pure photorealism, especially for portraits, hands, faces, and complex scenes. Flux is competitive on photoreal at lower cost but lacks the same fine-detail rendering Ultra produces.

Imagen 4 or Flux for creative control?

Flux. Seed control, image-to-image with adjustable denoising strength, inpainting, and outpainting all give you precise iteration on a single image. Imagen 4 is largely a one-shot generation API.

Flux at 5 credits per image is the cheapest tier on Viral Engine. Imagen 4 Standard is 8 credits, Fast is 8 credits, Ultra is 25 credits. For high-volume work, Flux is materially cheaper.

Open vs closed weights, does it matter?

For most users, no. The model runs server-side in both cases. For users who need to fine-tune or run locally, Flux is open-weights and accessible; Imagen 4 is closed-weights Google API. On Viral Engine you access both through the same dashboard.

Imagen 4 vs Flux: Which AI Image Model Should You Pick?

Two of the strongest image models in 2026 sit at opposite ends of the philosophy spectrum. Imagen 4 is Google's closed-weights flagship: maximum photorealism, three quality tiers, generated through Google's API. Flux is Black Forest Labs' open-weights model: maximum creative control, seed locking, image-to-image, inpainting, outpainting, at the cheapest credit cost on the platform.

Both are available on Viral Engine. This guide covers when to pick each one with cost math and real prompt examples.

The 30-second answer

Photorealistic hero shots and portraits: Imagen 4 Ultra.
Reproducible variations and seed-locked iteration: Flux.
Inpainting and outpainting: Flux.
High volume on a budget: Flux (5 credits per image).
In-image text and signage: Imagen 4.
Image-to-image with adjustable strength: Flux.

Side-by-side

Dimension	Imagen 4 Ultra	Flux
Photorealism (1-10)	10	8
Creative control	Low	High
Seed locking	No	Yes
Image-to-image	Limited	Yes (adjustable strength)
Inpainting	No	Yes (brush)
Outpainting	No	Yes (canvas expand)
In-image text	Best in class	Good
Cost per image	25 credits	5 credits
Speed (1-10)	5	7

Cost math: $14/month, what does it buy?

Essential plan, 4,000 credits:

Imagen 4 Ultra: 160 hero-tier images per month.
Flux: 800 images per month at 1K, ~530 at 2K, ~400 at 4K.

If you need 200+ images per month, Flux is the only economical choice. If you need 50 hero-tier images, Imagen 4 Ultra is the right pick.

When Imagen 4 wins

Anywhere the output is the deliverable, not a step:

Brand hero shots. Skin texture, surface accuracy, lighting fidelity. Imagen 4 Ultra's photorealism ceiling is the reason it exists.
Portrait closeups. Hands, faces, eye contact. The classic AI failure modes that Imagen 4 substantially fixes.
In-image typography. Logos, product labels, headlines. Quote the exact text in the prompt and Imagen 4 renders it cleanly.
Print and large-format. Detail holds up at scale; Flux can show artifacts when blown up beyond ~2K.

When Flux wins

Anywhere the workflow needs control or repeatability:

Seed-locked variations. Generate one composition you like, lock the seed, change small prompt details. Same composition, controlled variations. Imagen 4 has no equivalent.
Image-to-image refinement. Upload a draft, dial denoising strength, transform style or lighting while preserving composition.
Inpainting fixes. When 90% of an image is right and 10% needs surgery (a malformed hand, a wrong product label, a cluttered background corner). Brush mask, regenerate just that region.
Aspect-ratio expansion. Outpaint a portrait into a 16:9 banner without re-shooting.
Volume at low cost. 5 credits per image is half the price of Imagen 4 Standard.

The workflow we use

Internally, we don't pick one or the other; we chain them.

Generate 4 variations on Flux (cheap, fast, gives composition options).
Pick the strongest, lock the seed.
Refine the prompt and re-generate on Flux 2-3 more times until composition is dialed.
Take that final composition and feed it as image-to-image into Flux at low denoising strength to lock it in.
For the absolute final hero, regenerate the prompt on Imagen 4 Ultra. Pick the best.

Total cost per finished hero: ~50 credits (5 Flux runs at 5 credits + 1 Ultra at 25). Versus 100-200 credits if you generate Ultra-only from scratch hoping for a hit.

Prompt example: same brief, both models

Brief: Editorial portrait of a woman in her 40s, auburn shoulder-length hair, soft north-window light, emerald velvet backdrop, 85mm.

Flux output: Strong composition, accurate lighting, clean velvet texture. Slight softness on facial micro-detail. 5 credits, ~6 seconds.

Imagen 4 Ultra output: Clinical portrait quality, every pore and hair strand resolved, perfect eye reflection. Print-ready. 25 credits, ~12 seconds.

For a magazine cover, use Ultra. For an Instagram post, Flux is more than enough.

Open weights vs closed weights, does it matter?

For most users on Viral Engine, no. Both run server-side. You generate, download, ship. The open-weights distinction matters if:

You want to fine-tune the model on your own dataset (Flux supports this; Imagen 4 doesn't).
You want to run the model locally with no platform dependency (Flux runnable; Imagen 4 not).
You're sensitive to vendor lock-in concerns at the model layer.

Otherwise, both feel identical from inside the dashboard.

Bottom line

Pick Flux when you need control or volume. Pick Imagen 4 Ultra when output is the deliverable. Or do what we do: chain them. Cheap iteration on Flux, finishing pass on Ultra. The 70 free credits on signup let you test both before deciding.

More: Nano Banana 2 vs Imagen 4, prompt engineering guide, Imagen 4 deep dive, Flux deep dive