Repository navigation
第三個釋出版 / Third Release
這是第三個釋出版。
考慮了許久後,終於在這個版本做出一些較激進的改變:把本地 LLM 模式整個移除。
連帶地,也影響到了許多相關功能與設定。
以下簡單列出這個版本的主要更新:
1. 移除本地 LLM 模式
移除本地 LLM 模式,相關的設定功能也一併移除。
2. 大幅簡化設定介面
大幅簡化設定介面,把一些 API 模式不太需要的高階設定值通通移除。
同時,也移除了非高階使用情境下極少使用的設定項目。
主要原則就是讓這支 APP 更加傾向於純粹的 API 角色扮演聊天軟體,同時預設使用對象是非專業的純角色扮演玩家。
以下稍微列舉幾個主要移除項目:
Text Completion- 依關鍵字隱藏角色
- 通知功能
- 隱藏文字
- 停止序列 UI
- 專家設定頁
3. 新增主題顏色
新增幾個內建的主題顏色,目前總共提供 10 種主題。
有考慮過直接製作一個圖形化的顏色自訂功能,但這項工程太大,而且與聊天 APP 的核心主題沒有太大關聯,所以目前先暫時擱置。
4. 語音功能改成串流模式
語音功能改為串流模式(Streaming)。
這可以減少語音發聲時的停頓感。
經過我個人測試後,我認為改善相當明顯。
5. 新增「劇情作弊器」
新增了劇情作弊器(Story Cheat)。
這個功能可以在對話過程中直接下達影響回覆風格的指令。
其實本質上就是 Author's Note(作者筆記),只是我幫它取了一個比較有趣的名字。
在設計上也刻意保持簡單易用,不希望在裡面放入太多複雜的背景設定項目。
6. 新增動畫語音對嘴功能(測試用)
新增動畫語音對嘴功能(測試用)。
這個功能的觀賞性已經有了,但目前還不夠理想,所以暫時標註為測試用。
目前提供兩種對嘴模式:
LemonSlice
這是真正漂亮、生動的動畫對嘴效果。
角色會有非常自然的動態,反應速度也很快。
唯一的問題就是:非常貴。
這是一項以秒數計費的服務。
靜態圖片連播
這是比較陽春的方案。
它會生成原始圖片的三張嘴部變化圖,然後連續播放,藉此製造角色正在說話的錯覺。
整體表現方式類似一般 Anime 的嘴部動畫。
實際效果比我原本預期的還要好,但畢竟只有嘴巴在動,因此還是會存在一些僵硬感。
目前對嘴功能還不算理想。
我的目標是做出類似 D-ID 那樣,可以讓角色即時動態對嘴說話的效果。
但目前我還沒有找到一個能同時兼顧生動、快速,而且低成本的實現方式。
English Version
This is the third release.
After giving it a lot of thought, I finally decided to make some rather aggressive changes in this version: the entire Local LLM mode has been removed.
This change also affected quite a few related features and settings.
Here is a brief overview of the major updates in this release:
1. Removed Local LLM Mode
The Local LLM mode has been completely removed, along with all of its related settings and configuration options.
2. Significantly Simplified the Settings Interface
The settings interface has been significantly simplified.
Advanced options that are generally unnecessary for API-based usage have been removed, along with settings that are rarely used outside of advanced scenarios.
The main goal is to make this app more focused on being a pure API-based role-playing chat application, with non-professional role-playing users as the primary target audience.
Some of the major features and settings that have been removed include:
Text Completion- Hide characters by keyword
- Notification features
- Hidden text
- Stop Sequence UI
- Expert Settings page
3. Added More Theme Colors
Several built-in theme colors have been added.
There are now a total of 10 themes available.
I did consider adding a graphical color customization system, but the amount of work required would be quite large, while having relatively little connection to the core purpose of a chat application.
So for now, I've decided to put that idea aside.
4. Switched Voice Playback to Streaming Mode
Voice playback has been changed to streaming mode.
This helps reduce the noticeable pauses during speech playback.
After testing it myself, I feel that the improvement is quite significant.
5. Added "Story Cheat"
A new Story Cheat feature has been added.
It allows you to directly issue instructions during a conversation to influence the style and direction of the AI's responses.
Technically, this is essentially the same concept as an Author's Note, but I decided to give it a slightly more interesting name.
The feature is intentionally designed to remain simple and easy to use, rather than adding a large number of background or configuration settings.
6. Added Animated Lip Sync (Experimental)
An animated voice lip-sync feature has been added as an experimental feature.
The visual effect is already quite interesting, but it is still not ideal, which is why it is currently marked as experimental.
There are currently two lip-sync modes:
LemonSlice
This provides genuinely beautiful and lively animated lip-sync.
The character has very natural movement and reactions, while the response speed is also very fast.
The only problem is that it is very expensive.
The service is charged based on the number of seconds generated.
Static Image Animation
This is the more basic option.
It generates three different mouth variations based on the original image and plays them sequentially to create the illusion that the character is speaking.
The overall approach is similar to the way mouth animation is commonly handled in Anime.
The result is actually better than I expected, but since only the mouth is moving, there is still some noticeable stiffness.
The lip-sync feature is still not quite where I want it to be.
My goal is to create something similar to D-ID, where the character can speak with real-time, dynamic lip-sync animation.
However, I have not yet found an implementation that can simultaneously achieve natural-looking animation, fast response times, and low cost.