#I tested voice remixing on my voice clone, but I wish to change it back to what it was before. How

1 messages · Page 1 of 1 (latest)

grizzled bloomBOT
#
AI-generated Answer for lightbulb42_18597

Short answer: you don’t need to “revert.” Voice Remixing never edits your original clone—it creates a separate voice variant. Your original voice is still in My Voices. (elevenlabs.io)

What to do:

  • In the editor/TTS, select your original voice from My Voices and use that going forward. (elevenlabs.io)
  • Optional: remove the remixed variant. In My Voices, open the three‑dots menu for that voice and choose Delete voice. (help.elevenlabs.io)
  • API users: switch back to the original voice_id. Remix previews return a generated_voice_id that becomes a new voice when created; your original voice_id is unchanged. (elevenlabs.io)

Sources:
Voice remixing | ElevenLabs Documentation
Voice remixing | ElevenLabs Documentation
How do I delete voices from My Voices? – ElevenLabs
Remix a voice | ElevenLabs Documentation

signal iris
#

When I select the original voice, it only plays the remixed one. However, if I select a previous generation from my history, and click restore settings, the voice is correct. How can I make the voice the correct one when selecting it (instead of having to restore settings each time)

grizzled bloomBOT
#
AI-generated Answer for lightbulb42_18597

Short answer: Remixing never overwrites your original voice; it creates a separate voice variant. If selecting “the original” still sounds like the remix, you’re likely picking the remixed variant (or using paragraph-level overrides) in your project.

Do this to make the correct/original voice the one that plays on selection:

  • Verify which voice is the original vs the remixed variant in My Voices. Click More actions (three dots) → Copy voice ID, and use the original voice’s ID going forward. You can also rename voices via More actions → Edit voice to avoid confusion in the picker. (help.elevenlabs.io)
  • In your Studio project, reassign the original voice to your content globally rather than restoring paragraph-by-paragraph. Select the text range, choose the original voice from the Voices sidebar, then use the settings button next to that voice under “Used in Project” to apply settings across all paragraphs (or disable per‑paragraph “Override settings”). (help.elevenlabs.io)
  • If voice settings are causing unexpected timbre changes, reset/update the voice’s defaults (stability, similarity, style, speed, speaker boost). You can fetch defaults and edit per‑voice settings via the API if needed. (elevenlabs.io)

Why this works:

  • Remixing creates a new voice variant and does not change the original; you need to select the original voice asset/ID explicitly. (elevenlabs.io)
  • “Restore” in Generation History only reverts that paragraph; it doesn’t change which voice you’ll get next time you pick from the menu. Apply the voice and settings project‑wide instead. (help.elevenlabs.io)

If the original voice entry still outputs the remixed timbre everywhere (which shouldn’t occur per docs), please open a support ticket and include both voice IDs and a short screen recording so we can investigate. (help.elevenlabs.io)

Sources:
How do I find the voice ID of my voices via the website and API? – ElevenLabs
How can I change the voice and settings across multiple paragraphs in Studio? – ElevenLabs
Get default voice settings | ElevenLabs Documentation
Voice remixing | ElevenLabs Documentation
What is generation history in Studio? – ElevenLabs
How can I contact support? – ElevenLabs

signal iris
#

There are no other versions of that voice, when looking through my voices, so I'm not selecting the wrong voice.

grizzled bloomBOT
#
AI-generated Answer for lightbulb42_18597

Thanks for confirming. Per ElevenLabs’ documentation, Voice Remixing does not overwrite the original voice—remixing “creates a new voice variant,” and “your original voice remains unchanged and available in your library.” If you only see one entry in My Voices, that doesn’t match the documented behavior and should be reviewed by support. (elevenlabs.io)

What you can verify now

  • In My Voices, use More actions (three dots) → Copy voice ID to confirm exactly which voice you’re selecting. Each voice (original and any remixed variants) has a distinct voice_id. (help.elevenlabs.io)
  • Note that restoring a previous generation in Studio only reverts that paragraph to an earlier generation; it doesn’t modify or recreate the underlying voice asset in My Voices. (help.elevenlabs.io)

If the original voice isn’t visible as a separate entry

  • This contradicts the Voice Remixing docs, so please open a support ticket and include the voice_id you’re selecting, a brief timeline of when you remixed, and screenshots of My Voices. Use the contact form here. (help.elevenlabs.io)

Optional fallback

  • If you still have the original training audio, you can recreate the clone (results may vary slightly between clones). ElevenLabs does not allow exporting voice clones themselves. (help.elevenlabs.io)

If you’d like, I can outline exactly where to find and compare voice IDs in the UI steps.

Sources:
Voice remixing | ElevenLabs Documentation
How do I find the voice ID of my voices via the website and API? – ElevenLabs
What is generation history in Studio? – ElevenLabs
How can I contact support? – ElevenLabs
Can I export my voice clones? – ElevenLabs

signal iris
#

I see that if I restore settings and choose v3 alpha or create something in studio, then it chooses the remixed voice. When I change the model back to v2 turbo, the original voice is back

grizzled bloomBOT
#
AI-generated Answer for lightbulb42_18597

Thanks for the details—here’s what the docs say and how to lock the behavior you want:

What’s happening

  • Voice Remixing creates a new voice variant and does not overwrite the original voice. The remixed variant is “v3‑native” but is backward‑compatible with other models. (elevenlabs.io)
  • Studio supports multiple models (including v3) and you can change the model per project in Project settings. (elevenlabs.io)

How to get back to the original voice consistently

  1. Identify and select the original voice
  • In My Voices, copy the Voice ID for your original voice (More actions → Copy voice ID) so you can be sure you’re selecting the non‑remixed one in TTS and Studio. (help.elevenlabs.io)
  • If you see a separate remixed entry in My Voices and don’t want to use it, delete that variant (More actions → Delete voice). (help.elevenlabs.io)
  1. Set the model explicitly
  • In Studio: open Project settings and set the model you prefer (e.g., Turbo v2). Studio will then generate with that model for the project. (elevenlabs.io)
  • In API or advanced flows: specify both the model and the exact voice_id. For Turbo v2 use modelId eleven_turbo_v2; for v3 (Alpha) use eleven_v3. (help.elevenlabs.io)

Why v3 (Alpha) may sound like the remixed version

  • Remixed voices are produced as full‑quality v3 voices, so when you switch a project/model to v3 you may be hearing that v3‑native remixed variant if it’s the one assigned/selected. The original should still be available as a separate voice; selecting it (by name or Voice ID) should restore the original sound. (elevenlabs.io)

If the UI only shows one voice and it behaves like it’s been replaced (which would contradict the docs that remixing doesn’t overwrite the original), please send a short screen recording plus the affected Voice IDs to Support so we can investigate. (help.elevenlabs.io)

References

signal iris
#

It seems like when I use any other model than turbo v2, the voice clone becomes another voice, even though the voice id is the same

grizzled bloomBOT
#
AI-generated Answer for lightbulb42_18597

Short answer: yes—this can happen by design.

Why you’re hearing a different timbre on non‑Turbo v2 models

  • Voices can behave differently across models. Eleven’s docs note that “some voices may work better with one model… compared to another,” so the same voice_id is not guaranteed to sound identical on every model. (elevenlabs.io)
  • Your Professional Voice Clone (PVC) is automatically trained on specific models: Flash v2.5, Turbo v2.5 and Multilingual v2 (plus Flash v2/Turbo v2 if your training audio was English). You can optionally fine‑tune it onto additional models when they’re supported. If you pick a model your PVC hasn’t been fine‑tuned for, its rendering can differ. (help.elevenlabs.io)
  • Eleven v3 is in public alpha, with different behavior and higher variability; Eleven advises it requires more prompt engineering and is not yet tuned for real‑time reliability. That can make a voice feel “like another voice” relative to Turbo v2. (help.elevenlabs.io)

How to get back the “original” sound without restoring each time

  • Use a model your PVC is trained for (Turbo v2 or v2.5 in your case). That’s why switching back to Turbo v2 restores the expected timbre. (help.elevenlabs.io)
  • Save the voice’s settings on the voice itself (not just within a single generation). Set and persist high similarity_boost and enable use_speaker_boost, then adjust stability as needed. You can do this in My Voices or via the Edit voice settings API. (elevenlabs.io)
  • Be aware that “Restore” in Studio only restores the settings/audio for that paragraph’s generation; it does not globally change your voice’s saved settings. (help.elevenlabs.io)

If you want to use other models with the same timbre

  • In My Voices, hover your PVC and fine‑tune it onto any additional supported models (models with a “+” icon). You’ll be notified when new models support fine‑tuning. (help.elevenlabs.io)
  • If you experiment with v3 (Alpha), follow the v3 prompting guide and stability guidance to reduce divergence, but expect more variability than Turbo v2 for now. (help.elevenlabs.io)

Related references

  • Model IDs and selection (Turbo/Flash/Multilingual). (help.elevenlabs.io)
  • Default voices are trained across multiple models (which is why they tend to be more consistent model‑to‑model than custom PVCs). (help.elevenlabs.io)

If after saving voice settings and using a trained model you still get mismatches, follow the troubleshooting guidance or contact support. (elevenlabs.io)

signal iris
#

How do I train the voice on different models?

grizzled bloomBOT
#
AI-generated Answer for lightbulb42_18597

Short answer:

  • Per-model training is only available for Professional Voice Clones (PVC). Instant Voice Clones (IVC) cannot be fine‑tuned on specific models. (help.elevenlabs.io)

How to train a PVC on different models:

  • What’s automatic: Every PVC is automatically trained on Flash v2.5, Turbo v2.5 and Multilingual v2. If your PVC was trained on English audio, it’s also automatically trained on Flash v2 and Turbo v2. (help.elevenlabs.io)
  • Add/upgrade models: In My Voices, hover over your PVC’s name to see supported models. A tick means the voice is already fine‑tuned for that model; a plus means it’s available to fine‑tune—click the model to start. You’ll see progress on hover; due to caching you may need to refresh. You’ll get in‑app and email notifications when fine‑tuning completes. New fine‑tunable models will show an exclamation icon next to the voice. (help.elevenlabs.io)
  • Training time: After you start fine‑tuning, typical completion is roughly 2–6 hours; status is visible in My Voices and you’ll be notified when it’s ready. (help.elevenlabs.io)

Using the models after training (API):

  • Specify the target model via the model_id in TTS endpoints (e.g., eleven_turbo_v2, eleven_turbo_v2_5, eleven_flash_v2_5, eleven_multilingual_v2). You can list available models via the /v1/models endpoint. (help.elevenlabs.io)

If you’re using an Instant Voice Clone today and need consistent behavior across multiple models, you’ll need to upgrade that voice to a PVC to fine‑tune it per model as described above. (help.elevenlabs.io)

Sources:
What is the difference between Instant Voice Cloning (IVC) and Professional Voice Cloning (PVC)? – ElevenLabs
How do I add or upgrade the models used to train my Professional Voice Clone? – ElevenLabs
How do I add or upgrade the models used to train my Professional Voice Clone? – ElevenLabs
When will my professional voice clone (PVC) be ready? – ElevenLabs
How do I find the model ID? – ElevenLabs
What is the difference between Instant Voice Cloning (IVC) and Professional Voice Cloning (PVC)? – ElevenLabs

signal iris
#

Is it possible to train the voice for v3?