Which model to pick: a practical guide to the Visnea catalog
July 23, 2026
The catalog has more than twenty models, and the most expensive one is not automatically the right one. Selection tables for concrete tasks: product cards, poster text, reference edits, photo animation and video with audio.
Picking a model is not about "which is best overall" but "which is best for this task at this price". Below are the combinations we actually use. Prices are as of publication; the current price is always visible in the model list and the "Cost" line.
Images: a 10-second decision
| Task | Model | Why |
|---|---|---|
| Drafts, idea exploration | Z-Image (1 cherry) | Cheapest and fastest; built for iteration |
| All-round photorealism | Seedream 4.5, Seedream v4 (5 cherries) | Confident photographic frames, strong prompt adherence |
| In-image text, posters | Ideogram v3, GPT Image 2, Qwen Image | Precise typography; Qwen is the most affordable of the three |
| Reference-based edits | Nano Banana (3 cherries), FLUX Kontext | Edit an uploaded image from a prompt while keeping the scene |
| Consistent character across a series | Nano Banana, Nano Banana 2 | Hold the face and style between generations |
| Complex multi-step edits | Nano Banana Pro (from 12 cherries) | The flagship for precise edits and top quality, up to 4K |
| Maximum detail | FLUX.2 pro | The flagship for composition and detail |
| Speed/quality balance | FLUX.2 flex | Quality and speed controls per task |
| Stylized, bold looks | Grok Imagine | Expressive stylized frames from text and references |
The everyday chain: Z-Image to explore → Seedream 4.5 or FLUX.2 pro for the final → Upscale before print.
Video: pick by scenario
| Scenario | Model | Price reference |
|---|---|---|
| Cheap motion tests | Grok Imagine | 6 s at 480p — 6 cherries |
| Budget all-rounder | Seedance 1.5 Pro | 4 s at 480p — from 4 cherries |
| Animate a photo | Hailuo 2.3 / 2.3 Pro | 6 s at 768P — 15 / 23 cherries |
| Cinematic final | Kling 3.0 | 5 s — 34 cherries, with Pro and 4K modes |
| The fast Kling | Kling 3.0 Turbo | 720p/1080p, 3–15 seconds |
| Start → end frame transition | Kling 2.5 Turbo, WAN 2.7, Seedance 2.0 Mini | morphing between two images |
| One character across clips | Seedance 2.0 (up to 4 references) | references define the hero and style |
| Video with audio built in | Kling 2.6, Kling 3.0, Seedance 2.0 Mini, Gemini Omni Video | AI audio is a toggle and costs extra |
| Wide 21:9 cinema format | Seedance 2.0 / 2.0 Fast | plus an "Auto" mode matching your frame |
Mind the multipliers: seconds, quality and audio compound. Five seconds is almost always enough for social — do not pay for 15 out of habit.
Voiceover
For Russian text the catalog carries a fresh wave of TTS models — from fast to HD quality. Pricing follows text length (the per-block rate is shown in the form). A practical rule: record the voiceover after the final cut is approved, so you never pay to re-voice changed copy.
How to compare models yourself
- Take one benchmark prompt from your real work (not "a cat in glasses" — your actual product card).
- Run it on 2–3 candidates in the same frame format.
- Judge three things: prompt adherence, artifacts on hands/text/edges, usability without retouching.
- Lock in the winner — and switch models only when the task changes.
Catalog badges help you navigate: "New" marks fresh arrivals, "Pro" marks flagships, "Budget" marks the cheap entry point. The description under each model's name states plainly what it is good at — it is a cheat sheet, not marketing.
Frequent questions
Why does the same task come out differently on different models? Models have different temperaments: some are literal, some improvise. If a model invents too much, pick a more literal one (GPT Image 2, Nano Banana 2) and state the priority at the end of the prompt.
Should I always use the flagship? No. For a feed post, the difference between 3 and 18 cherries often disappears after the platform compresses the file. Flagships earn their price on print, close-ups, text and complex edits.
What if a model disappears from the list? The catalog is alive: weak models leave, stronger ones arrive. Your prompts carry over — the structure from our prompt guide works on any model.