What is Image generation?

Image generation Choose model and processing mode Cost, time, formats and parameters depend on the selected model.

Who is Image generation for?

Image generation Image generation Choose model and processing mode Cost, time, formats and parameters depend on the selected model.

Image generation · TelkNet

Image generationGPU workload

Image generation

Q: What tasks can Image generation help with?

Image generation; Available canvas sizes: 1024x1024, 1680x720, 1664x416, 416x1664, and 1024x1792.; Fast text-to-image exploration, moodboards, product concepts, editorial visuals, and creative drafts; Write a natural-language prompt with subject, composition, style, lighting, and any visible text in quotes. Current production deployment width: 256-1792 px, multiple of 16. Current production deployment height: 256-1792 px, multiple of 16.; Generate aesthetic directions, moodboards, and concepts from natural-language prompts; Use the Krea 2 Turbo official 8-step inference defaults for daily generation

Q: Which models and related terms are associated with Image generation?

Krea 2 Turbo, krea/Krea-2-Turbo, krea/Krea-2-Raw, krea-ai/krea-2, Krea 2 open-source release, Krea 2 technical report, Krea 2 prompting guide, Ideogram 4

5-1250 credits/useText promptEstimated 300-9900 seconds

Input

Write a natural-language prompt with subject, composition, style, lighting, and any visible text in quotes. Current production deployment width: 256-1792 px, multiple of 16. Current production deployment height: 256-1792 px, multiple of 16.

Output

Available canvas sizes: 1024x1024, 1680x720, 1664x416, 416x1664, and 1024x1792.

Use case

Fast text-to-image exploration, moodboards, product concepts, editorial visuals, and creative drafts

Choose model and processing mode

Choose by input and output needs, then comparable tests, published evaluations and capabilities. Missing scores do not imply lower quality; more parameters or a newer release do not prove superiority. This is a suggested usage order, not an overall leaderboard.

Cost, time, formats and parameters depend on the selected model.

Under Maintenance

This tool is under maintenance. Please try again later.

Models, parameters, and sources

Krea 2 generates images from natural-language prompts. TelkNet uses the Krea 2 Turbo checkpoint with official inference defaults, focusing on fast aesthetic exploration, moodboards, product concepts, and editorial visuals; Raw is not exposed for daily generation.

Krea 2 Turbokrea/Krea-2-Turbokrea/Krea-2-Rawkrea-ai/krea-2Krea 2 open-source releaseKrea 2 technical reportKrea 2 prompting guideIdeogram 4Ideogram 4 toolideogram-ai/ideogram-4-nf4Krea 2GitHub release

Krea 2 Turbo

Krea 2 Turbo is the official 8-step distilled open-weight checkpoint Krea recommends for daily inference.

krea/Krea-2-Turbo krea/Krea-2-Raw krea-ai/krea-2 Krea 2 open-source release Krea 2 technical report Krea 2 prompting guide

Krea 2 Turbo official inference settings are 8 steps, CFG 0.0, and mu 1.15.
The release describes a 12B dense DiT with Qwen Image VAE and Qwen3-VL text encoder components.
Krea 2 Turbo is an open-weight release under the Krea community license.
The official prompting guide recommends natural-language prompts, detailed composition, and quoting rendered text; Krea is better for fast aesthetic exploration than for replacing Ideogram 4 structured typography and layout control.

Ideogram 4 NF4

Ideogram 4 NF4 is the 9.3B open-weight text-to-image foundation model path used here for design, typography, structured prompts, bbox layout control, and native 2K generation.

ideogram-oss/ideogram4 Ideogram 4 model page Ideogram 4 technical blog ideogram-ai/ideogram-4-nf4 ideogram-ai/ideogram-4-fp8

The official model card describes Ideogram 4 as an open-weight image model at the frontier of design, and ranks it first among open-weight models on Design Arena.
Current weights: TelkNet uses ideogram-ai/ideogram-4-nf4. This is the official public NF4 quantized weight path, not a full-precision checkpoint; an FP8 variant is also available for evaluation.
Prompt control: the official path supports structured JSON prompts, visible text, bbox coordinate layout, color, and style controls for posters, brand work, packaging, and product visuals.
Text strength: the official model card emphasizes multilingual text rendering, signage, logos, titles, watermarks, and multi-line text as core strengths of Ideogram 4.
Official highest-quality recommendation: V4 Quality 48 with an explicit canvas up to 2048x2048.

Model comparison

Model

Role

Why it is used

Output impact

ModelKrea 2 Turbo

krea/Krea-2-Turbo

RoleDaily inference path

Why it is usedKrea 2 Turbo official inference settings are 8 steps, CFG 0.0, and mu 1.15.

Output impactPNG image candidates from the Turbo checkpoint with the official 8-step settings.

ModelIdeogram 4

Ideogram 4 tool ideogram-ai/ideogram-4-nf4

RoleTypography and design comparison

Why it is usedIdeogram 4 can be better for exact text rendering, logos, packaging, bbox layouts, and poster typography.

Output impactUse the quick link when visible text and layout control are more important than general image Elo.

ModelIdeogram 4 nf4

ideogram-ai/ideogram-4-nf4 ideogram-oss/ideogram4

RoleDeployed local open-weight model

Why it is usedHigh-resolution PNG images generated through the official Ideogram 4 NF4 open-weight path

Output impactPNG image

ModelIdeogram 4 fp8

ideogram-ai/ideogram-4-fp8 ideogram-oss/ideogram4

RoleOfficial alternate quantization

Why it is usedListed for model-family transparency; not the current TelkNet run path.

Output impactReference variant

ModelOther open-weight image models

ideogram-ai/ideogram-4-nf4 Ideogram 4 technical blog

RoleExternal model context

Why it is usedThe technical blog describes the 9.3B single-stream DiT architecture, Qwen3-VL text encoder, and structured prompting design.

Output impactRanking context

Paper and benchmark notes

This checkpoint has no reproducible public score table. See the linked release material for details.

Krea 2 Turbo

Paper and benchmark notes

Krea 2

GitHub release - 2026

The Krea 2 repository documents the Raw open-weight checkpoint, inference guidance, and local usage.

krea-ai/krea-2 Krea 2 Turbo Krea 2 Raw Technical report

Krea 2 open-source release

Krea release page - 2026

The Krea release page describes the Raw open-weight release and architecture components.

Release page Krea 2 Turbo Krea 2 Raw Prompting guide

Ideogram 4

Hugging Face model card - 2026

The official Ideogram 4 model card documents NF4/FP8 variants, benchmarks, structured prompting, 2048 support, and recommended sampler settings.

ideogram-4-nf4

Ideogram 4.0 Technical Details

Ideogram technical blog - 2026

The technical blog describes the 9.3B single-stream DiT architecture, Qwen3-VL text encoder, and structured prompting design.

Technical blog Official inference repository ideogram-4-fp8 Model page

Official / repositories

krea/Krea-2-Turbo krea/Krea-2-Raw krea-ai/krea-2 Krea 2 open-source release Krea 2 technical report Krea 2 prompting guide ideogram-oss/ideogram4 Ideogram 4 model page Ideogram 4 technical blog ideogram-ai/ideogram-4-nf4 ideogram-ai/ideogram-4-fp8

Papers / technical notes

No public paper link is listed; use the official repo or runtime implementation as the source.

Parameter guide

Prompt: natural-language description; quote any text that should appear in the image.
Width: this production deployment exposes 256-1792px in multiples of 16; official Turbo sampling remains 8 steps, CFG 0.0, and mu 1.15.
Height: this production deployment exposes 256-1792px in multiples of 16, aspect ratio from 1:6 to 6:1; the tested stable pixel budget is 960x1680.
Steps: blank uses the official Turbo recommendation of 8.
CFG: blank uses 0.0 for Turbo, disabling classifier-free guidance.
Mu: blank uses the official Turbo recommendation of 1.15.
Seed: 0 uses TelkNet's stable automatic seed; click the dice for a random explicit seed.

Use cases

Generate aesthetic directions, moodboards, and concepts from natural-language prompts
Use the Krea 2 Turbo official 8-step inference defaults for daily generation
Switch to Ideogram 4 when typography, logos, packaging, or exact poster layout is the priority

Workflow

1Write a natural-language prompt and choose a supported canvas size
2Use the Krea 2 Turbo checkpoint with official inference defaults
3Download every generated PNG from the completed task

Before you submit

Put visible text in quotes and describe layout plainly
Use the Turbo defaults for steps, CFG, and mu unless you need an explicit override
Credits are recalculated from pixels, steps, and candidate count

Related use-case guides

Working on a similar task? These guides list the matching tools, model boundaries, sources, and FAQs for each scenario.

AI image generation · online tool and model guide Krea 2 image generation · online tool and model guide ChordEdit image editing · online tool and model guide

Quick feedback

Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.