The Halloween Invitation

14 models were given the same prompt, and the community voted blind on which outputs looked best. How it works

While rendering simple text is no longer a real challenge for today’s SOTA models, this test shows something more interesting: how much visual taste a model has, and whether it can create a layout that feels like it came from a professional designer instead of a basic Canva template (no offense to Canva).

Blind Vote This Challenge

#1 — Seedream 4.0

Prompt

“Vintage gothic Halloween party invitation. Dark parchment poster, spooky border with webs and thorns, central glowing jack-o-lantern, bats, twisted trees, moody night sky. Add elegant gothic title text saying "Halloween Party Invitation", a small scroll banner saying "You are invited to a night of frights", and event details at the bottom: Date: 30.10.2026 Time: 7pm Location: The Arches, NYC Spooky but polished, cinematic lighting, square format.”

Voters were asked to judge by Visual Taste Text Accuracy Layout

Challenge Rankings

14 models

#	Model	Price	¢/img	Speed	Elo
1	Seedream 4.0 ByteDance	$$	3¢		1164
2	Nano Banana 2 Google	$$	2.2¢		1164
3	FLUX.2 [klein] 4B Black Forest Labs	$	0.1¢		1152
4	FLUX.2 [dev] Turbo fal	$	0.8¢		1143
5	Nano Banana Pro Google	$$$	6.7¢		1126
6	Qwen Image 2.0 Alibaba	$$	3.5¢		1099
7	FLUX.1 [dev] Black Forest Labs	$$	2.5¢		1095
8	GLM-Image Z.ai	$$$	5¢	—	1081
9	Seedream 4.5 ByteDance	$$	4¢		1081
10	Qwen Image 2.0 Pro Alibaba	$$$	7.5¢		1079
11	Nano Banana Google	$$	3.9¢		1074
12	DALL-E 2 OpenAI	$$	1.6¢		1063
13	Stable Diffusion 3.5 Medium Stability AI	$$	2¢		1056
14	FLUX.1 [schnell] FP8 Black Forest Labs	Free	Free		984

Seedream 4.0 and Nano Banana 2 share the top spot at 1164 Elo, demonstrating that mid-tier priced models currently offer the best balance of complex layout design and text precision. The 4B-parameter FLUX.2 [klein] serves as a significant outlier, securing third place (1152 Elo) while costing 95% less and generating images three times faster than the leaderboard leaders.