AI media tools according to AI: ElevenLabs, HeyGen and Veo are near-unanimous, image generation is split
Ten models were asked which AI video, image, voice, avatar, music and clipping tools to use. ElevenLabs and HeyGen were named in every answer and first in 94% and 92%. Image generation has no leader.
Figure 01 · Edition 2026-09
AI video generation
How often each product was named, and how often it appeared first.
- VeoNamed: 100%Named first: 56%
- RunwayNamed: 92%Named first: 28%
- KlingNamed: 80%Named first: 4%
- SoraNamed: 70%Named first: 4%
- HeyGenNamed: 48%Named first: 0%
- SeedanceNamed: 44%Named first: 8%
- SynthesiaNamed: 38%Named first: 0%
- LumaNamed: 32%Named first: 0%
Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09
Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.
Key findings
ElevenLabs: 100% of voice answers, first in 94%. HeyGen: 100% of avatar answers, first in 92%. The two strongest positions in the Index.
Veo leads video generation with 100% share and 56% first; Runway is second at 92% and 28%. Sora was named in 70% of answers and first in 4%.
Image generation has no leader: Adobe Firefly 96% share and 36% first, Midjourney 96% and 34%. OpenAI's own models most-named Adobe Firefly, not GPT Image.
Suno leads music at 90% and 64% first. Udio was named in 66% and first in none.
Opus Clip leads clipping at 96% and 72%. Synthesia was named in 100% of avatar answers and first in 4%.
ElevenLabs is named first in 94% of AI voice answers and HeyGen in 92% of avatar answers, the two most decisive results across all 33 categories. At the other end, image generation is a coin flip between Adobe Firefly and Midjourney, and OpenAI’s own models did not favour OpenAI’s image model.
AI video generation
Veo in 100% of answers, first in 56%. Runway 92% and 28%. Kling 80% and 4%. Sora 70% and 4%. HeyGen appeared in 48% of video answers as an adjacent avatar tool. Seedance was named in 44%. All four providers agreed on Veo, including OpenAI’s three models, which named Google’s product above their own Sora. Reddit was the top cited source with 128 citations, the highest Reddit count in the Index.
AI image generation
Adobe Firefly 96% share and 36% first. Midjourney 96% and 34%. Ideogram 70% and 0%. GPT Image 68% and 6%. Recraft 54% and 8%. Nano Banana 36% and 10%. The providers split: OpenAI’s three, Google’s two, Perplexity’s two and Claude Fable 5 most-named Firefly; Claude Opus 5 most-named Midjourney; Claude Sonnet 5 most-named GPT Image. The category has no default, and Adobe’s commercial-safety positioning is what tips the models when they have to choose.
AI voice generation
ElevenLabs 100% and 94%. Murf 74% and 0%. Descript 72% and 4%. Play.ht 50%. Google 36%. No other category has a leader named first this often. Murf is the clearest case in the Index of a vendor on every list and never the pick.
AI avatar video
HeyGen 100% and 92%. Synthesia 100% and 4%. Colossyan 72% and 0%. D-ID 64% and 4%. Tavus 30%. Synthesia and HeyGen are named in every answer; the models pick HeyGen 46 times and Synthesia twice. HeyGen’s own site was the top cited vendor domain at 100 citations. Descript was never named for avatars.
AI music generation
Suno 90% and 64%. Udio 66% and 0%. Soundraw 56% and 22%. Beatoven and AIVA 50% each. Mubert 48%. Providers split at the edges: Claude Fable 5 and both Perplexity models most-named Soundraw, the other seven Suno. Udio at 66% with zero firsts is the second-biggest mention-to-pick gap in the media set after Murf. Riffusion and Splash were never named.
Short-form clipping
Opus Clip 96% and 72%. Descript 74% and 4%. Vizard 64% and 6%. CapCut 54% and 6%. Riverside 50%. Choppity 38% and 10%, with its own site the top cited vendor domain at 95 citations. Sonar Pro alone most-named CapCut. Podcastle, Crayo and 2short were never named.
What the field shows
Where a product defined its category, the models know it: ElevenLabs, HeyGen, Suno, Opus Clip. Where the big labs compete with each other, the answer is contested and the models do not favour their maker. OpenAI’s models put Veo above Sora and Firefly above GPT Image. Google’s models put Suno above their own Lyria, which did not register.
YouTube was the top source in five of six media categories; Reddit led video generation.
What this means for a vendor
The media categories reward the product that became the noun. A second-place product with 70% to 100% share and single-digit firsts (Synthesia, Murf, Udio, Sora, Runway) is one the models respect and do not recommend. The gap is specific to the reason given in the first sentence of the answer.
Limitations
Fifty answers per category. Model-vendor overlap (OpenAI, Google) is measured, not adjusted for. Media tools change fast; this is one month. One edition.
Explore 5 more category charts
Figure 02 · Edition 2026-09
AI image generation
How often each product was named, and how often it appeared first.
- Adobe FireflyNamed: 96%Named first: 36%
- MidjourneyNamed: 96%Named first: 34%
- IdeogramNamed: 70%Named first: 0%
- GPT ImageNamed: 68%Named first: 6%
- RecraftNamed: 54%Named first: 8%
- Nano BananaNamed: 36%Named first: 10%
- CanvaNamed: 36%Named first: 0%
- Stable DiffusionNamed: 32%Named first: 0%
Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09
Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.
Figure 03 · Edition 2026-09
AI voice generation
How often each product was named, and how often it appeared first.
- ElevenLabsNamed: 100%Named first: 94%
- MurfNamed: 74%Named first: 0%
- DescriptNamed: 72%Named first: 4%
- Play.htNamed: 50%Named first: 2%
- GoogleNamed: 36%Named first: 0%
- SpeechifyNamed: 28%Named first: 0%
- OpenAINamed: 26%Named first: 0%
- WellSaidNamed: 26%Named first: 0%
Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09
Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.
Figure 04 · Edition 2026-09
AI avatar video
How often each product was named, and how often it appeared first.
- HeyGenNamed: 100%Named first: 92%
- SynthesiaNamed: 100%Named first: 4%
- ColossyanNamed: 72%Named first: 0%
- D-IDNamed: 64%Named first: 4%
- TavusNamed: 30%Named first: 0%
- CreatifyNamed: 24%Named first: 0%
- DeepBrain AINamed: 24%Named first: 0%
- CaptionsNamed: 20%Named first: 0%
Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09
Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.
Figure 05 · Edition 2026-09
AI music generation
How often each product was named, and how often it appeared first.
- SunoNamed: 90%Named first: 64%
- UdioNamed: 66%Named first: 0%
- SoundrawNamed: 56%Named first: 22%
- BeatovenNamed: 50%Named first: 6%
- AIVANamed: 50%Named first: 0%
- MubertNamed: 48%Named first: 2%
- ElevenLabsNamed: 28%Named first: 0%
- Stable AudioNamed: 16%Named first: 0%
Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09
Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.
Figure 06 · Edition 2026-09
Short-form clipping tools
How often each product was named, and how often it appeared first.
- Opus ClipNamed: 96%Named first: 72%
- DescriptNamed: 74%Named first: 4%
- VizardNamed: 64%Named first: 6%
- CapCutNamed: 54%Named first: 6%
- RiversideNamed: 50%Named first: 2%
- ChoppityNamed: 38%Named first: 10%
- KlapNamed: 26%Named first: 0%
- SubmagicNamed: 20%Named first: 0%
Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09
Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.
Behind the figures
Method and sources
Edition 2026-09: 5 fixed buyer prompts per category, sent to 10 models through their APIs. Each category contains 50 recorded answers. Answer share is the fraction of answers naming a vendor. Named first records the vendor's position before other tracked vendors, not the strength of a recommendation. This is one dated sample; small differences can reflect answer variation. API responses can differ from consumer chat products. Read the full method.
- AI video generation: recorded answers 50 answers · Edition 2026-09
- AI image generation: recorded answers 50 answers · Edition 2026-09
- AI voice generation: recorded answers 50 answers · Edition 2026-09
- AI avatar video: recorded answers 50 answers · Edition 2026-09
- AI music generation: recorded answers 50 answers · Edition 2026-09
- Short-form clipping tools: recorded answers 50 answers · Edition 2026-09
Cite this research
Memetik. AI media tools according to AI: ElevenLabs, HeyGen and Veo are near-unanimous, image generation is split. Memetik, 2 September 2026. Data: edition 2026-09. https://www.memetik.ai/research/ai-media-tools-according-to-ai
Charts and figures: CC BY 4.0. Credit Memetik Index and link to the source.